Download Foretelling The Future Of Prognostication: A Historically

Yale University
EliScholar – A Digital Platform for Scholarly Publishing at Yale
Yale Medicine Thesis Digital Library
School of Medicine
January 2012
Foretelling The Future Of Prognostication: A
Historically Inspired Domain-Based Approach For
The Elderly
John Thomas
Yale School of Medicine, [email protected]
Follow this and additional works at: http://elischolar.library.yale.edu/ymtdl
Recommended Citation
Thomas, John, "Foretelling The Future Of Prognostication: A Historically Inspired Domain-Based Approach For The Elderly" (2012).
Yale Medicine Thesis Digital Library. Paper 1770.
This Open Access Thesis is brought to you for free and open access by the School of Medicine at EliScholar – A Digital Platform for Scholarly
Publishing at Yale. It has been accepted for inclusion in Yale Medicine Thesis Digital Library by an authorized administrator of EliScholar – A Digital
Platform for Scholarly Publishing at Yale. For more information, please contact [email protected].
FORETELLING THE FUTURE OF PROGNOSTICATION: A HISTORICALLY
INSPIRED DOMAIN-BASED APPROACH FOR THE ELDERLY
A Thesis Submitted to the
Yale University School of Medicine
in Partial Fulfillment of the Requirements for the
Degree of Doctor of Medicine
by
John Michael Thomas
2012
FORETELLING THE FUTURE OF PROGNOSTICATION: A HISTORICALLY
INSPIRED DOMAIN-BASED APPROACH FOR THE ELDERLY
John M. Thomas and Terri R. Fried. Section of Geriatrics, Department of Internal
Medicine, Yale University School of Medicine, New Haven, CT.
Hypothesis: 1) Physicians’ decisions to discuss hospice as an option for terminally ill
patients are based on a limited approach to prognostication that excludes many patients
who may benefit from discussions. 2) Identifying broader domains of health most
important for prognostication, as an alternative to calculating life expectancy or mortality
risk, might encourage prognostication and improve physician-patient communication.
Aims: 1) To examine the association between physicians’ prognostic assessments and
their discussion with patients about hospice. 2) To identify the domains of health-related
characteristics of older hospitalized patients and nursing home residents most strongly
associated with short-term mortality.
Methods: Following an historical introduction on prognostication, we describe two
empiric studies. First, we performed secondary analyses of surveys administered to 215
patients age ≥60 years with advanced cancer, chronic obstructive pulmonary disease, or
heart failure that were performed at least every 4 months for up to 2 years, as well as
surveys to their respective physicians at least every 6 months. Then we performed a
systematic review of prospective studies that evaluated the association between at least
one health-related patient characteristic and mortality within one year among patients age
≥65 years. All studies published in English in MEDLINE, Scopus, or Web of Science
before August 1, 2010 were eligible. We categorized the characteristics into a series of
domains. Using the results of multivariable analyses, we ranked domains within each
study according to strength of association with mortality, then calculated the overall
relative strength of each domain as compared to other domains across studies.
Results: Apart from diagnosis of cancer, the factors most strongly associated with
hospice discussion in our empiric analysis were physicians’ estimate of and certainty
about patient life expectancy (P<0.001). That said, physicians did not anticipate the
deaths of 40% of patients. In the systematic review, we classified characteristics
associated with mortality from forty-eight studies into seven domains: cognitive
function, disease diagnosis, laboratory values, nutrition, physical function, pressure sores,
and shortness of breath. The most important domains for prognostication were nutrition
and shortness of breath among general nursing home residents; physical function and
shortness of breath among nursing home residents with dementia; disease diagnosis,
nutrition, and pressure sores among hospitalized patients for in-hospital mortality; and
physical function and nutrition among hospitalized patients for mortality up to one year.
Conclusions: Clinicians’ discussion of hospice for patients with advanced illness relies
largely on a highly unreliable prognostic approach that involves estimated life
expectancy, and many clinicians whose patients might benefit from learning about
hospice are not having these discussions. Among a large number of health-related
characteristics of older persons shown to be associated with short-term mortality, a few
consistently important domains provide broad, easily measurable factors that may
promote an approach to prognostication that simply alerts physicians to patients who are
at increased risk for mortality, rather than aiming for certainty in life expectancy, thus
encouraging physician-patient communication for elderly persons nearing the end of life.
ACKNOWLEDGEMENTS
I thank Dr. Leo Cooney for his invaluable contributions to the systematic review,
and Dr. Margaret Drickamer for her review of the final draft of this thesis. I also thank
Melissa Grafe for her assistance with historical texts, and Jan Glover and Lei Wang for
their assistance with designing the electronic literature search. A portion of this research
was supported by the James G. Hirsch, MD Endowed Medical Student Research
Fellowship.
“Life is short, the art long,
opportunity fleeting,
experience treacherous,
judgment difficult.”
–Hippocrates, Aphorisms
TABLE OF CONTENTS
Historical Introduction…………………………………………………….7
Statement of Purpose………………………………………………………32
Part I: Prospective Cohort Study...............................................................33
Methods……………………………………………………………. 34
Results……………………………………………………………... 39
Discussion…………………………………………………………..45
Part II: Systematic Review………………………………………………. 49
Methods……………………………………………………………. 51
Results………………………………………………………………62
Discussion…………………………………………………………...79
Overall Conclusion…………………………………………………………86
References…………………………………………………………………..88
7
Historical Introduction
The purpose of this historical exercise, which will be followed by two empiric
studies, is not so much to provide a comprehensive account of the thought and practice of
prognostication throughout Western medicine, because that would require a great deal
more discussion; but rather, to trace the evolution of a word – prognosis – and thus
support the argument that this venerable and sacred practice of prognostication has lost
over time its true appeal and meaning; for the field of medicine, if it truly seeks progress,
must look to the past as it looks to the future.
What is the reason for this need to approach medicine, and thus prognostication,
with an awareness of historical practice? Does medicine not forge ever onward in its
pursuit of knowledge and its ability to heal? The answer lies in the fact that medicine,
ever-advancing scientifically, underwent a paradigm shift that directed attention towards
individual diseases and away from diseased individuals, resulting in both tremendous
advancements and harmful neglects. In Ancient times and for long after, when the
mechanisms of disease had not yet been worked out, an emphasis was placed on the
broader manifestations of disease, such as signs and symptoms, for determining
prognoses. While this approach may seem oversimplified to the modern reader, it was
the best that could be done at the time and served a valuable and cherished purpose when
treatments were largely not available. On the other hand, the practice of modern
medicine, which understands the histopathological basis for disease, has placed an
emphasis on disease diagnosis for determining prognoses. This approach may, too, be
oversimplified, especially for elderly patients with comorbid conditions and for whom
8
treatments often do more harm than good. Although in place for decades, this recent
paradigm emphasizing disease diagnosis has become increasingly problematic.
Two Medical Traditions
Although throughout Western history the practice of medicine has varied greatly,
appreciating the larger theoretical structures that have informed the practice of medicine
over time is essential for understanding the transformation that occurred in
prognostication. Indeed, an argument can be made for two largely distinct medical
“traditions,” separated by a scientific revolution alluded to previously, that have operated
under the same name of medicine but with decidedly different theoretical bases and
emphases.
The first thread of medical “tradition” can be traced from Classical Antiquity until
approximately 1800, in which learned physicians fundamentally relied on the writings of
their predecessors, recent or distant, in their understanding of health and disease.
Scientific knowledge handed down from antiquity went for a long time unquestioned, and
even as advances and discoveries were made that sometimes contradicted previous
thought, the basic underpinnings of medicine remained the same.1 A principle example
of this is the persistence of the theory of humors, from which the practice of bloodletting
was derived.
In the nineteenth century, discoveries were made that forever changed the face of
medicine.2 The acceptance of germ theory brought with it fundamental insights into the
mechanisms of disease. Great strides followed, including developments in vaccination
and antiseptic practices. A glance at the past exposed glaring errors, and thus the former
centuries-old “tradition” of medicine no longer retained its importance or centrality.
9
While occasional references to the writings of esteemed physicians of the past continued
to be made, especially quotes attributed to Hippocrates, such references tended to serve
as general “wisdom” in the art of medicine rather than having immediate authority. The
rapid advancements of science seemed to cast a shadow over the art of medicine, which,
while still present, struggled to adapt to an ever-changing practice as developments
exploded into the twentieth century and beyond.
The Decline of Prognosis
An article on prognostication in Lancet from 1934 reads: “Of the three great
branches of clinical science – diagnosis, prognosis, and treatment – prognosis is
admittedly the most difficult. It is also that about which least has been written and of
which our knowledge is least systematized.” 3 This observation is typical of the second
“tradition” of Western medicine, in which a precise understanding of the basis of disease,
aided by technological progress, resulted in dramatic advancements in diagnostics and
treatment options. The practice of prognosis, on the other hand, did not enjoy the same
excited attention. For example, one can see the gradual disappearance of discussions
about prognosis from medical textbooks in the twentieth century, whereas during the first
“tradition” prognosis was often heralded as something to be cultivated. Although in
recent times accurate prognostication has become the target of increased scientific
inquiry, attempts to encourage its clinical use have been unsuccessful, especially for
elderly persons nearing the end of life. A more thorough discussion of this problem will
appear later.
The explanation I would suggest for the decline in status of prognosis is threefold,
although essentially a unified concept: first, a transformed understanding of disease
10
directed attention away from individual patients and towards individual diseases, thus
equating prognosis with diagnosis and destroying the tie between prognosis and broader
patient characteristics; second, the rise in effective treatments for individual diseases
further diminished the need to assess prognosis for individuals with these diseases; and
third, prognostication has been perceived as characteristic of the first medical “tradition,”
which was dismissed by modern science.
Ultimately, in the face of an aging population, physicians would do well to realize
that the interplay of comorbid conditions and the often-mixed effects of treatments
necessitate the use of signs and symptoms once again, in addition to diagnostics, for the
purpose of prognostication. In other words, a careful consideration of both individual
diseases and diseased individuals is required for the restoration of prognosis to the same
status as diagnosis and treatment.
In the subsections that follow, prognostication as it has been described in various
historical periods will be discussed, with occasional comments about concurrent
developments in medicine to provide context.
Classical Antiquity
The body of works attributed to Hippocrates contains several treatises on
prognostication, and among these is On Prognostics, which contains an excellent
overview of the importance of this art in the Ancient practice of medicine: “It appears to
me a most excellent thing for the physician to cultivate prognosis; for by foreseeing and
foretelling, in the presence of the sick, the present, the past, and the future, and explaining
the omissions which patients have been guilty of, he will be the more readily believed to
be acquainted with the circumstances of the sick.”4 Notably, the definition of prognosis
11
rendered here is broader than the modern accepted usage. Rather than being limited to
predicting the likely course and outcome of illness, it also involves relating to the patient
current and prior aspects of that patient’s illness. The reason for this becomes clear in his
subsequent discussion: that the purpose of prognostication is as much about gaining the
trust and confidence of patients as it is about effecting a cure. “Thus a man will be the
more esteemed to be a good physician, for he will be the better able to treat those aright
who can be saved, from having long anticipated everything; and by seeing and
announcing beforehand those who will live and those who will die, he will thus escape
censure.” The ability to prognosticate, then, was an important means by which
physicians could distinguish themselves, demonstrate their worth, and avoid criticism.
To the modern reader, the Hippocratic use of the word prognosis might be
perceived as heavily contingent upon the concept of diagnosis. After all, arriving at the
correct diagnosis in modern times generally affords an explanation of the patient’s past
and present state of health, as well as a prediction for the future. It must be remembered,
however, that the understanding of disease in Classical Antiquity was based not on an
understanding of “agents of disease” or cellular pathology, but rather on a long-accepted
theory of four humors, in which an imbalance of one or more humors would result in a
disease state. Thus, prognostication in the time of Hippocrates was rather contingent
upon observing the signs and symptoms of illness, and placing them in the context of
attributes of the patient, such as age and robustness, to arrive at insights about the past,
present, and future course of illness.
Though Hippocrates probably enjoys the greatest renown today among Ancient
practitioners of medicine, the central figure in the development of the first medical
12
“tradition” was probably Galen, who lived nearly five centuries after the time of
Hippocrates. Galen’s influence was so great that subsequent generations accepted his
writings without question. His writings not only documented his own experimental
findings, which were considerable in volume, but summarized and synthesized the
writing of his predecessors, not the least of which was Hippocrates. Because of the
tremendous regard for his work in the centuries that followed, Galen was in effect the
“sieve” through which previously known medicine passed onto subsequent generations.1
Likewise, with regards to prognostication, his treatise Of Critical Days
importantly discusses in great detail certain theories that not only indicate some of the
practices of the time but would exert a lasting influence.5 In this treatise, he discusses the
concept of crisis, which is the crucial point at which a sick patient either dies or is
restored to health. He also describes a method for calculating critical days, or the days in
which a crisis is most likely to occur. This method relies entirely upon the principles of
astrological medicine, in which the sun and moon were believed to influence the course
of illness, for good or bad, in predictable intervals in accordance with their movements
with respect to the Earth. Galen’s description of astrological technique as a
prognostication tool would have influences well into medieval times.
From Medieval to Modern Europe
A great deal of medical texts from Classical Antiquity would have been lost in the
centuries that followed, had they not been preserved as part of the Arab-Islamic medical
tradition.1 Indeed, until the late eleventh century, very little changed in the practice of
medicine in Europe. It was not until the translation of medical texts from Arabic and
13
Greek, augmented by the rise of universities, that the tradition begun by Hippocrates and
extended by Galen and Arab medicine was introduced into the Latin-speaking world.
One result of this sharing of texts is the Summary on Crisis and the Critical Days,
a medical work on prognostication written by an anonymous author in the late thirteenth
century.6 It effectively marks the introduction into Medieval Europe of the complex and
highly sophisticated astrology described by Galen, as it heavily relies on Of Critical Days
for its material and frequently cites it. The Summary on Crisis and the Critical Days
achieved wide popularity among university physicians in the Latin-speaking world over
the course of three centuries.
In the sixteenth and seventeenth centuries, a pronounced renewal of interest in
Galen and Greek medicine occurred in Europe. Such an interest can be observed in the
English text, “Prognostication drawen out of the books of Ipocras, Avicen, and other
notable auctours of physycke,” printed at least as early as 1550 by Robert Wyer.7
Noticeably, the use of the word prognosis in this text is limited to predicting “whether in
peryl of death be in them or not, the pleasure of almyghty God reservyd.” An example of
its approach to prognostication can be found in the discussion on sweat, here spelled
“swete”: “Where the swete is, there is the [sickness]. The swete which cometh now and
then is nought. Swete with [weakness] of the pulse, is nought. Note swete in the [head],
in the continual fever onely is nought. And if it be colde, it betokeneth death.” It is
unclear for whom this particular publication was intended, but given that in early modern
Europe learned medicine was often popularized in the vernacular and used by educated
lay people as well as trained physicians (an English statute in the mid-sixteenth century
14
even legalized the unlearned practice of medicine), it is possible that Wyer published the
text for a less discerning population.
Among university-trained physicians, probably the most outstanding work on
prognostication in the sixteenth and seventeenth centuries was called The presages of life
and death in diseases: in seven books, written by Prosper Alpini.8 This text relies heavily
on the Hippocratic and Galenic writings for their descriptions of the methods of
prognostication, achieving a comprehensive synthesis. The author claims to have
confirmed these methods “not only by the sentiments and opinions of the ancient
physicians, but also by a long course of attentive observation and experience.” In fact,
nearly one hundred pages are devoted to a discussion of critical days and crises, including
a description of the astrological origin of the concept of critical days. Alpini’s use of the
term prognostication, unlike that of Wyer, is consistent with the Hippocratic definition
found in On Prognostics.
At the same time that a renewed interest in Classical Antiquity occurred in the
sixteenth and seventeenth centuries, which comprised part of the early modern era in
Europe, various challenges against conservatism developed and ran alongside it.1 The
problems of plague and syphilis, which seemed to be “contagious,” forced flexibility into
the Galenic approach that did not allow for “agents” of disease to be communicated from
one person to another. New discoveries were made in anatomy and physiology,
including quite influentially the concept of circulation of the blood. Finally, the
introduction and acceptance of modern philosophy promoted a “new science,” which
reduced biological processes to mere mechanical events, removing the mysterious,
cosmological component allowed for by Aristotelian philosophy. Overall, it was
15
increasingly acceptable, even fashionable, to reject Galen’s teachings. In contrast to
Galen, however, the reputation of Hippocrates was relatively unscathed. While Galen
was accused of going beyond experience and thus dogmatically espousing erroneous
teachings, Hippocrates was seen as guided by experience and personifying great bedside
observation.
The resultant transformation of medical ideology that well over a century of “new
science” incited may be seen in a treatise by James Harvey written in 1706, called,
Praesagium Medicum, or, the Prognostick Signs of Acute Diseases.9 In the text, signs
and symptoms are discussed in exquisite detail with respect to their further
characterization, likely cause, and the likelihood of recovery. Although descriptions of
the mechanisms of disease are often reminiscent of humoral theory, Harvey decidedly
discounts the approach of Classical Antiquity in his discussion of crises and critical days,
proclaiming the progress of medicine in its improved knowledge of prognostication:
“Everybody knows how religiously critical days were observed by the Ancient
physicians. But, later ages have wiped off the dust of antiquity, discovered its infirmities,
and enriched the art of physick with closer observations and discoveries…” He
specifically discounts astrological medicine, stating that crises and critical days were “not
fixed to a certain and determinate number, the moon’s motion, or that of any other
constellation…” His understanding of crises and critical days, rather, was based upon the
inner workings of the body without regard for the influence of the cosmos.
Nevertheless, Harvey’s use of the word prognosis remains consistent with the
Hippocratic definition. In addition to stating the likely outcome associated with
particular signs and symptoms, he characterizes them in detail and discusses their causes.
16
In fact, the minority of the discussion is devoted to predicting the future course of illness.
Instead, Harvey emphasizes the importance of not being too specific in prognostication:
“A prudent and wary physician therefore will be moderate and ambiguous in his
promises, and reserved in his prognostics, unless founded upon certain and infallible
signs.”
It would seem that Harvey’s approach to prognostication is the result of a keen
awareness of the Hippocratic writings combined with a heightened scrutiny and
skepticism towards certain concepts passed down from Classical Antiquity. His
reluctance to offer predictions about the future outcome of illness is perhaps a
characteristic of the “new science” that rejected the orthodox explanations of crises and
critical days and acknowledged an uncertainty about the timing of recovery or death.
Nonetheless, he appreciated the scope of prognostication as defined by Hippocrates and
discussed fully the aspects of prognostication where he felt the science had advanced.
This advancement was limited, however. Despite the accumulation of greater knowledge
during the sixteenth and seventeenth centuries, only rarely did effective new treatments
develop. Instead, the focus of scientific progress was on anatomy, physiology, and the
mechanisms of disease.
Although the eighteenth century still saw no radical transformation in medical
therapeutics, the Enlightenment movement strengthened physicians’ ambitions to further
medical “progress” through scientific findings, with the ultimate aim to gain better
control over nature.1 Certain strategies, including quantification and classification,
allowed for greater objectivity and a more systematic approach to understanding health
17
and disease. The future loomed large as the science of medicine progressed, although
major breakthroughs were still lacking.
Characteristics of the eighteenth century approach to medicine can be found in the
treatise, Observations on the prognostic in acute diseases, written by Charles Le Roy and
published in English in 1782.10 For instance, there is an attempt to achieve greater
precision in prognostication: “In a work of this sort, only by much the smaller number of
prognostics, can be as certainties: the rest will vary in their degrees of probability. It has,
therefore, been my aim, to adapt my expressions to the degree of probability, which
seemed to belong to each prognostic.” While the treatise is a tribute to Hippocrates,
making frequent reference to him, Le Roy emphasizes the need to verify the observations
handed down from Antiquity: “I have very seldom spoken of any prognostics, that I have
not seen confirmed in my own practice…”
In the organization of the treatise, and in the use of the word prognosis, Le Roy
takes very much the same approach found in the Hippocratic writings. An important
underlying assumption is that particular signs and symptoms, even if they may be caused
by a variety of different diseases, still portend the same prognostic significance and can
be addressed in the same way regardless of the underlying disease.
The Nineteenth Century
In the aftermath of the political revolutions of the late eighteenth century, Western
society was becoming increasingly fluid, with a growing emphasis on the rights of
individuals.2 This was the milieu in which profound change and upheaval occurred in
medicine. On the one hand were physicians who looked to the future and saw “tradition”
as something to be avoided; on the other hand were physicians who held fast to
18
“tradition,” even all the way back to Hippocrates, and were suspicious of so-called
scientific “advancements.” Some saw the time-tested method of acquiring knowledge
through clinical observation at the bedside as being threatened by the growth of
experimental science. This tension makes Pasteur’s achievements as a non-physician
scientist all the more remarkable, as he helped define a revolution in medicine through
his advocacy of germ theory and his developments in the fields of microbiology and
immunology. Although new scientific discoveries often met with resistance and only
gradually won general acceptance, the underpinnings of health and disease were
profoundly changing.
In The Sequels of Disease, written in 1896, Dyce Duckworth offers some insights
into the dramatic changes occurring in medicine and their impact on the practice of
prognostication: “I venture to state that…in spite of the extraordinary advances made in
all branches of the sciences…the attention of physicians has been somewhat inadequately
directed to the subject of prognosis in diseases.”11 In fact, Duckworth was unaware of
any other work in the nineteenth century purely dedicated to the subject of prognosis. He
asserted that so much attention in medicine was being devoted to acquiring accurate
factual knowledge that few physicians had taken the time to synthesize and reflect upon
the information so that broader observations could be made.
In lamenting the endangered status of prognostication, Duckworth observes a key
distinction: “It was in connection with a humoral pathology that prognosis made the
greatest progress, and achieved its highest triumphs. The doctrines of solidism on the
other hand have uniformly proved inimical to the study of prognosis.” It is unclear
exactly what he means by this, because he does not go on to explain why “solidism,” the
19
doctrine that cells and tissues are the focus of disease, is not conducive to
prognostication.
One can observe, however, that The Sequels of Disease is organized in its
discussion of prognosis according to disease rather than by sign and symptom. Thus, the
discussion becomes quite different: there are no assumptions about particular signs and
symptoms having the same prognostic significance regardless of disease, as was the case
in Le Roy’s work and all the works on prognostication prior to it back to the time of
Hippocrates. This new approach reflects the revolution that had occurred in the
understanding of pathogens as agents of disease. The science of prognostication, then,
was to have a different foundation, and the writings on prognostication prior to the
revolution, including the Hippocratic works, would seem less relevant than ever.
The Twentieth Century
The tumultuous era that encapsulated the two World Wars was characterized by a
steady growth in medicine.2 Hospitals grew in size, specialization increased, medical
technologies such as x-ray machines and electrocardiographs came into widespread use,
and penicillin was developed and successfully aided Allied troops in the later years of
World War II. After the war, the field of medicine developed at an unprecedented rate,
with immense increases in research funding and extensive drug development. The strong
awareness of the past that had characterized the practice of medicine throughout the
entire “first” tradition had essentially vanished, as physicians no longer looked to the
writings of Hippocrates. An ever-increasing body of knowledge and technological
capability was seen as affirmation that progress was being achieved. By the 1970s,
however, the public increasingly began to question medical authority, as healthcare
20
expenses kept escalating and occasional mishaps like the thalidomide tragedy occurred.
The surge of new treatment options began to slow by 1980, and it became clear that cures
for cancer and other diseases like Parkinson’s would be slow in development. Infectious
disease, once thought to be nearly a thing of the past, began to reassert itself with drugresistance; the identification of HIV/AIDs in particular extinguished hopes of eliminating
infection. Despite increasing treatment capabilities and financial investments in
medicine, the ironic phenomenon of “doing better and feeling worse” arose in the
developed West.
Evidence for the decline in attention to prognostication over the course of the
twentieth century can be seen quite simply in the evolution of medical textbooks.
Nicholas Christakis demonstrated this through a content analysis of a series of editions of
The Principles and Practice of Medicine, a textbook authored originally by William
Osler.12 From 1892 to 1988, the percentage of chapter lengths devoted to prognosis
decreased steadily from about 10% to 0%. He observed that as the discussion of
treatment on a topic increased, the discussion of prognosis inversely decreased,
presumably because as disease became more treatable the likely outcome became more
routine. For highly treatable diseases, then, the need for prognostic considerations was
diminished.
In the same article, Christakis argued that the status of prognostication also
suffered in the context of an evolving concept of disease during the “Oslerian era.”
Whereas prior to the time of Osler disease was considered to be highly personal and
greatly influenced by the “constitution” of the patient, in the “Oslerian era” a shift
occurred towards focusing on the agents of disease. In this way, two very dissimilar
21
patients infected by the same pathogen could be seen as having the “same” illness, with
characteristics of the individuals relegated to the background.
Perhaps a third explanation for the decline of the status of prognostication at the
turn of the century was its perception as “unscientific.” In The Evolution of Modern
Medicine, for example, Osler discusses medical knowledge and practice as it developed
from the time of the Classical Antiquity to the present, and the only mention of
prognostication in the entire volume is in the context of the astrological practices of
medieval physicians.13 Although Osler indeed devoted portions of his medical textbook,
The Principles and Practice of Medicine, to discussions of prognosis, a close inspection
of the final edition written solely by him reveals high variability in whether prognosis is
discussed for any given disease.14 To cite a few examples, the discussion of prognosis
for lobar pneumonia is extensive, for typhus fever it is limited to an estimated mortality
rate, and for Hodgkin’s disease it is absent.
Despite the general decline in attention paid to prognostication over the course of
the twentieth century, the topic never disappeared entirely from academic discussion, and
even some attempts to foster the general practice of prognostication occurred. The start
of the twentieth century seems to have been, at least in part, a time for retrospection. An
editorial from the Journal of the American Medical Association (JAMA) in 1901 assures
readers that prognostication “is not entirely a natural gift, or a mysterious power granted
to the few, but that it is just as much the result of reading, study, observation at the
bedside and in the deadhouse, and of logical reasoning as is a scientific diagnosis or
rational therapy.”15 One physician, in 1904, referring to the legacy of Hippocrates,
writes, “The older prognosis, far from being negligible, is really of fundamental
22
importance. It gives the accumulated experience of ages that, untrammeled by detail,
carefully noted broad and elementary features of disease.”16
In the decades that followed, discussions about prognostication seem to have
shifted away from historical contextualization towards improving the science. From
1934 to 1936, an entire series of articles dedicated to disease-specific discussions on
prognostication were published in Lancet.3 In 1953, a physician describes
prognostication as a “stepchild in medical advance,” and insists that “we can do better
than we have in the past, or are doing now, by approaching the subject more scientifically
and seriously.”17
That same year, the first prognostic index was developed for patients admitted to
the hospital with a myocardial infarction.18 While the original purpose was to develop a
way to quantify patient severity of illness for the purpose of comparing the equivalency
of the experimental and control arms of an experiment, this “Pathologic Index Rating”
was found to be closely related to mortality rate. Following this initial paper, countless
other prognostic indices were developed covering a multitude of patient populations, with
newer indices sometimes offering advantages over older ones. An editorial in the New
England Journal of Medicine in 1971, on the calculation of prognosis, states, “Properly
derived, a prognostic index can guide the physician in his discussion with the family. It
can provide a basis for evaluating the consequences of medical care and for assessing
therapeutic innovation.”19
Clearly, as evidenced by the exponential development of prognostic indices from
the 1950’s onward, a promising approach had arrived, one that would ensure the
advancement of the science of prognostication. Would this development finally restore
23
prognosis to the same status in medicine as diagnosis and treatment? The likely answer
could be found in an in-depth analysis of a landmark study.
The SUPPORT Experiment
The largest-scale attempt to improve prognostication through a research
intervention was the Study to Understand Prognoses and Preferences for Outcomes and
Risks of Treatment (SUPPORT), conducted in the mid-1990s.20 The basis for the
intervention was the initial observational finding that discussions and decision making
between physicians and patients regarding end-of-life care uncommonly happened
substantially before death. Only half of physicians knew when their patients preferred
not to have cardiopulmonary resuscitation (CPR), and half of do not resuscitate orders
were written within 2 days of death. Furthermore, patients were found to commonly
receive aggressive treatment near the end of life: half of patients in their final
hospitalization spent at least 8 days in the intensive care unit (ICU), on a mechanical
ventilator, or in a comatose state. Finally, pain control was largely inadequate, given that
half of able-to-communicate patients reported moderate to severe pain most of the time in
their final days.
Based upon the largely unfavorable data on the care patients received approaching
the end of life, an intervention was designed to address these deficiencies by providing
physicians with accurate prognostic information regarding individual patients, including
the likelihood of mortality each day up to 6 months and predictions of future functional
ability, as well as providing physicians with patient preferences for end-of-life care. A
prognostic model was derived from a population of seriously ill patients who had one of
the following diseases: acute respiratory failure, multiple system organ failure with
24
sepsis or malignancy, coma, chronic obstructive pulmonary disease, congestive heart
failure, cirrhosis, colon cancer, and lung cancer.21 The steepness of the prediction curve
used for any given patient was based on the class of disease, and the placement of the
curve with respect to the x- and y- axes was based on severity of disease, as calculated
from 14 variables easily obtained from hospital records. Thus, the model generated a
point estimate of the likelihood of surviving to each day shown on the graph, along with
error bars to show variability in these estimates. A unique prediction could be made for a
given patient on every reporting day. This model was shown to be about as accurate in
its estimates as attending physicians were, but combining the estimate of the model with
that of the attending resulted in a much improved estimate over either alone.
The intervention aspect of SUPPORT utilized this model in an attempt to improve
the parameters of patient care described above through the provision of timely and
accurate prognostic information to physicians and patients, the elicitation of patient
preferences for end-of-life care, and the facilitation of communication and planning by a
skilled nurse. Despite all these efforts, the experiment was a failure in that it did not
improve any of the target outcomes although the study was sufficiently powered to detect
small differences. The authors were at a loss to explain the reasons for this. Many
hypotheses were generated about potential limitations of the study design, but the
ultimate consensus was that none of these limitations could account for the utter lack of
change in any measure. Questions were asked about whether physicians and patients
even agreed that these measures of end-of-life care were in need of improvement.
Perhaps there was nothing wrong with the design of the trial itself, it was suggested, but
rather more systemic changes were required before real changes could occur.
25
Clearly, the failure of SUPPORT was of such epic proportions that one could
reasonably suggest the trial’s inability to win the hearts of the physicians involved. The
task of relaying the prognostic information itself was not a problem, given that reports
were automatically provided to physicians and that nurses were hired to discuss prognosis
with patients. However, the provision of this information, even in an environment that
encouraged discussions, did nothing to change the approach of physicians.
One is tempted to ask about the attitudes of physicians towards the provision of
this meticulously calculated but highly statistical prognostic information. Did it actually
change their perceptions about the patient’s life expectancy? After all, estimates of the
model were only as good as the physician’s best guess. The additional consideration of
combining estimates of the model with physicians’ estimates to result in a more accurate
prediction at first seems encouraging, but it offers a mixed message to physicians that
makes it unclear whether the model can even be trusted, and if not trusted, exactly how
much the prediction should be altered. Given that discussions were suggested to be
explicitly based on the calculated prognostic information provided, how would
physicians explain to patients such alterations in a way that inspires confidence?
Physicians’ reservations about offering prognostic estimates to patients were later
characterized in a national survey of internists, in which they described prognostication
as being “stressful” and “difficult,” and believed that patients expect too much certainty
and may lose confidence in them if an incorrect prediction is made.22 Perhaps most
revealingly, 90% of physicians believed they should avoid being too specific in their
prognostic estimates, which calls into question the very approach of the SUPPORT trial.
Thus, evidence suggests that physicians were probably less than enthusiastic about being
26
offered calculated prognostic estimates and had misgivings about discussing them with
patients.
In addition to questioning the approach of offering calculated prognostic estimates
to physicians and patients in SUPPORT, one might also wonder about the lack of
improvement in communication between physicians and patients. The proportion of
patients reporting a discussion about CPR was no different from that of the control group,
even though over 40% of patients for whom CPR was not discussed reported that they
wished to discuss it. Nor was the proportion of patients reporting a discussion about
prognosis with their physician different from that of the control group, although again
over 40% of patients not reporting this discussion said they wished to have it. Granted,
physician reporting of these discussions was not obtained, allowing for the possibility
that patients forgot about conversations or did not recognize them as such; however, such
an influence would be expected to affect both the experimental and control groups
equally, thus maintaining whatever differences may have resulted from the intervention.
Therefore, it seems that while many patients wanted to have these discussions, their
physicians were either unaware of it or unwilling to honor the wishes of their patients.
Clearly the problem of lack of communication was rooted in more than simply the lack of
readily available prognostic estimates.
The SUPPORT trial is the strongest scientific demonstration available that the
meticulous provision of prognostic calculations that are as accurate as can be reasonably
achieved does not improve communication between physicians and patients, and has no
effect on decisions about the aggressiveness of care at the end of life. Supported by
evidence from physician survey, there are strong reasons to assert that such calculations
27
cannot achieve enough accuracy to make physicians comfortable with these sensitive
discussions of prognosis for patients approaching the end of life. The provision of
detailed estimates that are no better than physicians’ estimates and that could perhaps be
improved by physicians’ estimates introduces a complex psychological game for having
end-of-life discussions that is challenging at best.
The Legacy of Prognostic Models for Seriously Ill Patients
Two years after the publication of the primary results of SUPPORT, Joanne Lynn
and others published a paper in a low impact journal with data on the limited ability of
prognostic models to identify patients near the end of life.23 Two different models were
analyzed for the relationship between median estimates of survival and actual time to
death: the SUPPORT model, whose population is described above, and the APACHE III
model, whose population consisted entirely of ICU patients.
In the SUPPORT model, the median predicted chance of survival for 2 months
was 0.51 (0.31-0.66) 1 week before death, with substantial variations according to disease
diagnosis. The APACHE model was slightly less optimistic for the same parameter (0.45
median predicted chance of survival).
The overall conclusion was that in order to make plans about care and properly
support patients and families, discussions need to happen while the patient still has a
decent chance of surviving the current episode of illness, and that designations of
“terminally ill” for the sake of public policy would be necessarily arbitrary and
controversial.
Into the Twenty-First Century: Prognostic Indices for Clinical Practice?
28
The future of prognostication in both clinical research and clinical practice has
been unclear for well over a decade. Overall, it would seem that the lessons from the
SUPPORT trial were either ill-acknowledged or quickly forgotten: the failure of
SUPPORT was followed by a surge of new prognostic models, many of which attempted
to identify patients at risk for mortality within 1 year or less. Typically, the information
provided in indices generated from these models was less extensive than the information
provided to physicians and patients in SUPPORT, and prognostication based on these
indices required calculation on the part of the physician.
An editorial on the use of prognostic indices in clinical practice appeared in
JAMA in 2001.24 The authors observed that while prognostic indices are widely
developed in clinical research, their use in clinical practice is surprisingly rare. Various
challenges in the development and use of prognostic indices were discussed, including
the greater difficulty in generating life expectancy estimates as opposed to mortality risk
estimates, the difficulty of memorizing the elements of a risk index and calculating a
score, the lack of data comparing model estimates against clinician estimates, the lack of
explicit advice for applying estimates to clinical decisions, the “fatalism” implicit when
factors included in prognostic indices are unmodifiable, and the difficulty patients
sometimes have in comprehending probabilities. Overall, the authors declared that “the
strongest argument for prognostic indices is that they facilitate professional
communication,” although such communication of probabilities among physicians is
unlikely to dominate the medical field in the near future.
An evaluation of selected mortality prediction tools, published in 2011,25
demonstrated that most tools have only modest accuracy, with large variation in
29
discriminating performance when compared in different clinical studies. Even tools that
have found wide clinical use, such as APACHE II and MELD score, had inconsistent
accuracy when tested in different studies. It was suggested that existing literature may
even be biased toward exaggerating the accuracy of these tools since studies with
unimpressive results are less likely to be reported or published. Additionally, most
mortality prediction tools lack demonstrated clinical utility.
Prognostication for the Elderly
There has been a recent surge of interest in evaluating prognostic models designed
for elderly populations. The concurring opinion is that the use of such models in clinical
practice is premature. For instance, one systematic review identified a total of 193
models consisting of participants 50 years and older, and found that only 34% of these
models were externally validated, and only 2% were validated in more than two studies.26
Another systematic review identified 16 prognostic indices generated from patient
populations whose average age was 60 years or older and that predicted absolute risk of
mortality from 6 months to 5 years.27 It was found that none of the indices had excellent
discrimination (C statistic of ≥0.90), and only 2 indices were externally validated by
researchers other than those who developed the index. The authors concluded that while
prognostic indices are potentially useful for influencing clinical decisions that rely on
estimated life expectancy, their use cannot be recommended before further studies
demonstrate their accuracy in diverse populations and their ability to modify clinical
outcomes.
An editorial accompanying the latter systematic review discussed the importance
of prognostication for making fully informed clinical decisions.28 It argued that elderly
30
populations, who are more likely to have competing comorbidities and decreased life
expectancy, can especially benefit from the use of prognostication. For instance,
estimates of life expectancy can be directly applied to clinical guidelines as outlined in
the systematic review; some examples of decisions related to considerably diminished
life expectancy, in addition to the hospice eligibility guidelines, include discontinuation
of statins with life expectancy of ≤6 months, and non-operative management of
asymptomatic abdominal aortic aneurysms with life expectancy of <1-2 years. The
editorial also discussed a number of problems with current prognostic indices. In
addition to being premature for clinical use, most indices are designed to generate
mortality risk for a specific time interval rather than estimates of life expectancy; the
latter approach would be more clinically applicable. Also, there is the problem of the
lack of evidence that prognostic indices can improve patient outcomes. Despite these
limitations, the editorial argued that clinicians ought to be trained currently in how to use
prognostic tools.
The Future of Prognostication for the Elderly
Despite the greater accumulation of prognostic indices for elderly persons over
the past two decades, a substantial amount of evidence calls into serious question whether
such methods will ever achieve enough accuracy to have more than limited usefulness in
the clinical setting. Even if one or more indices eventually manages to gain acceptance
by clinical research standards, based on accuracy in diverse populations and improvement
of clinical outcomes in randomized controlled trials, it has yet to be shown that doctors
will be willing to use these inherently imperfect tools in prognostication for individual
patients. One might expect that if these prognostic indices are ever applied to specific
31
clinical decisions, the higher the stakes of the decision, the less likely doctors will be to
use them. For example, they would be especially problematic for identifying elderly
persons nearing the end of life for the purpose of discontinuing curative treatments.
What is the underlying problem with prognostic indices? Do they not at least
consider more than individual diseases in generating prognostic estimates? Perhaps the
problem lies in the quantification itself, which not only may give a false impression of
accuracy, but also represents a population-based average rather than truly representing an
individual.
The future of prognostication for the elderly lies in realizing that certainty is
unachievable, especially for persons with comorbid illness and for whom treatments are
often problematic. Instead of aiming for an unachievable goal, clinical researchers might
seek to use the substantial amount of available scientific evidence in a way that considers
broader manifestations of disease, such as signs and symptoms, in addition to disease
diagnosis, while deemphasizing quantitative prognostic estimates. This approach draws
upon historical perspectives on prognostication, while simultaneously maintaining the
scientific rigor that has long characterized clinical research.
32
STATEMENT OF PURPOSE
Hypotheses:
1) Physicians’ decisions to discuss hospice as an option for terminally ill patients
are based on a limited approach to prognostication that excludes many patients who may
benefit from discussions.
2) Identifying broader domains of health most important for prognostication, as an
alternative to calculating life expectancy or mortality risk, might encourage
prognostication and improve physician-patient communication.
Aims:
1) To examine the association between physicians’ prognostic assessments and
their discussion with patients about hospice.
2) To identify the domains of health-related characteristics of older hospitalized
patients and nursing home residents most strongly associated with short-term mortality.
33
PART I: Prospective Cohort Study
While attention to prognostication in medicine has declined over the past century,
so that it no longer enjoys a status alongside diagnosis and treatment as it once did, it has
regained some consideration in recent years with the development of an ever-increasing
number of prognostic models. That being said, the concept of prognosis has narrowed
significantly since the first medical “tradition,” now referring almost exclusively to
quantitative estimates of life expectancy or mortality rate. Perhaps the most classic
example of the modern use of prognosis is the Medicare Hospice Benefit requirement of
a life expectancy of ≤6 months for hospice enrollment,29 although notably alterations to
the eligibility criteria have been suggested in an attempt to more reliably identify persons
who would be appropriate for hospice.30 Proposed characteristics to identify such
persons include functional status, quality of life, and burden of symptoms rather than
estimates of life expectancy.
Hospice referral is simply one issue among many that physicians face as they
attempt to guide patients who are approaching the end of life. Another issue is informing
patients about their potential care options, including hospice services. In fact, one
identified barrier to hospice use is physicians’ lack of discussion about hospice with the
patient and family. 31,32 This observation has been supported by retrospective studies, in
which caregivers, recalling their conversations with physicians, often deny any
communication about hospice or alternative treatment options.33-35 Such discussions
about treatment options may be vital for guiding patients in their transition from being
gravely ill to dying.36
Although a few studies have examined determinants of patient referral to
34
hospice,37-39 no prospective study has examined determinants of discussions about
hospice, regardless of whether the patient ultimately utilizes hospice services. Thus, it is
unclear what role prognostication has in physicians’ decisions to discuss the option of
hospice with seriously ill patients.
The aim of this prospective cohort study40 was to examine patient-and physicianrelated factors associated with reported hospice discussions, with a particular emphasis
on the relationship between physicians’ prognostic estimates and hospice discussions.
METHODS
The data set used for this first component of the study was created by Dr. Fried
and detailed elsewhere.41 A description of participants and data collection contributing to
the data set appears below.
Participants
Participants of the study were ≥60 years of age and had a primary diagnosis of
cancer, chronic pulmonary obstructive disease (COPD), or heart failure (HF). Persons
screened for the study were being cared for as inpatients in a Veterans Affairs hospital, a
university teaching hospital, and a community hospital; as outpatients in two Veterans
Affairs hospitals; and as outpatients in four oncology, three pulmonology, and six
cardiology practices in the greater New Haven area. Each of the participating hospitals’
human investigations committees approved the study protocol. All patients gave
informed consent.
Sequential medical charts were reviewed for the primary eligibility criterion of
advanced illness, as defined by the National Hospice Guidelines42 or those criteria used in
35
the SUPPORT trial.43,44 A second eligibility criterion was the need for assistance with
one or more instrumental activities of daily living (IADLs)45; this was determined by
telephone screening and chosen to improve the identification of persons with advanced
illness,46 Patients were required to be full-time residents of Connecticut and have no
cognitive impairment as evaluated by a test of executive functioning47 and the Short
Portable Mental Status Questionnaire.48 Screening and enrollment were stratified
according to diagnosis in order to achieve equal numbers of participants with a diagnosis
of cancer, COPD, and HF.
Of the 548 persons identified by medical chart review, 30 were not contacted
because their physicians declined permission, 24 died prior to telephone screening, 19
declined screening, and 6 could not be reached. Of the persons screened by telephone,
108 were excluded because they were independent in all IADLs, 77 because they were
cognitively impaired, and 6 because they were residents of a state other than Connecticut.
Of the 279 eligible persons, 51 declined participation and 2 died before enrollment. The
final sample consisted of a total of 226 persons, with 82% participation of eligible
persons. Participants and non-participants did not differ statistically according to gender,
age, or Charlson comorbidity index score,49 with P<0.05 as cutoff. Eight percent of
eligible persons with HF declined participation, compared to 19% with cancer and 25%
with COPD (P=0.02). Of the 226 participants, eight (4%) withdrew after the initial
interview, 26 (12%) died before a follow-up interview, and three (1%) were not able to
complete follow-up interviews. Of the 124 participants still living one year after
initiation of the study, 98 (79%) agreed to participate for the second year.
36
Patient participants identified the physician primarily responsible for the
management of their primary diagnosis. Of 105 physicians identified, 96 (91%)
consented to participate and completed interviews for a total of 215 patients. Patient
participants whose physician participated in the study did not differ from patients whose
physician did not participate, when compared according to age, gender, ethnicity,
education, and income. None of the physicians of patients with cancer declined to
participate, compared to 15% of physicians for patients with HF and 2% of physicians for
patients with COPD and (P<0.001). The database used for this study included only the
215 patients whose physicians participated.
Data Collection
Patients were interviewed in their homes every four months for up to two years
and immediately following any decline in status, defined by the presence of one of the
following: need for assistance with an additional activity of daily living,50 hospitalization
for ≥7 days or resulting in discharge to a nursing home or rehabilitation center, or
enrollment in a hospice program. Physicians filled out a survey by mail every six
months. For this study, we used the last completed physician survey for each respective
patient. To ensure correspondence of information, we used the patient interview that
most closely preceded the physician survey.
The outcome variable was whether the physician discussed hospice. We
determined hospice discussions by physicians’ answer to the question of whether they
had discussed hospice with the patient and/or family. If physicians reported not
discussing hospice, they were asked to choose from a list of reasons why. We
determined receipt of hospice services by patient self-report, which was supplemented by
37
surrogate report if the patient was too ill to participate in an interview or died during the
study.
Descriptive and analytic variables extracted from patient interviews included
measures of health, sociodemographic, and psychosocial status. We dichotomized
ordinal variables at clinically relevant cut points. Health status variables included selfrated health (“excellent,” “very good,” “good,” “fair,” or “poor”) and an assessment of
symptoms, using the Edmonton symptom assessment scale.51 Sociodemographic
variables included gender, age, education, ethnicity, marital status, living arrangement,
and sufficiency of monthly income.52 Psychosocial variables included perceived
prognosis (patients were asked, “If you had to take a guess, how long do you think that
you might have to live?,” with responses of “<1 month,” “1-6 months,” “7-12 months,”
“13-23 months,” “2-5 years,” “6-10 years,” or “>10 years”), overall quality of life (“best
possible,” “good,” “fair,” “poor,” or “worst possible”),53 willingness to undergo major or
minor therapies if they would restore the patient’s current state of health (major therapies
were described as “being in the intensive care unit, receiving surgery, or having a
breathing machine” and requiring hospitalization of “at least a month,” while minor
therapies were described as “[receiving] intravenous antibiotics and oxygen,” and
hospitalization of “a few days to a week”), and awareness of alternatives to
hospitalization (patients were asked a series of questions: “If your illness should become
worse than it is now, what, if anything, has your doctor told you about how you could be
treated?,” then, “If you were sick enough that you potentially would need the hospital, do
you think that you would have any choices other than being hospitalized?,” and if so,
“What is/are the choices?,” followed by, “If you wanted to stay out of the hospital, do
38
you know of any services that could help you?,” and if so, “What are they?”). Each
interview contained the health status and psychosocial variables described above.
Physician surveys included the following descriptive and analytic variables: best
prediction of the patient’s life expectancy (<1 month, 1-6 months, 7-12 months, 13-23
months, 2-5 years, 6-10 years, or more than 10 years), and level of certainty regarding it
(99%, >90% certain, 50-90%, 10-49%, <10%, or <1% certain); whether they had told
the patient that he or she could die as a result of the disease; and whether they had
discussed with the patient his or her life expectancy.
Statistical Analysis
Statistical analyses were performed by Mr. O’Leary. Patient-reported and
physician-reported variables were described using frequencies and proportions, with the
variables expressed in total and stratified according to diagnosis. Bivariate analysis was
used to examine the association between these variables and physician report of hospice
discussion, by means of the chi-square test or, in the case of small cell sizes, the Fisher
exact test. Variables associated in bivariate analysis with hospice discussion (P<0.15)
were included in a logistic regression model (gender was entered into the model a priori).
Given that physicians’ estimates of life expectancy and discussions of hospice were
strongly associated, we examined the accuracy of physicians’ estimates by calculating the
frequency of these estimates among the subset of patients who died during the study.
39
RESULTS
Patient Population
Patient-related characteristics are shown in Table 1. Although 71% of patients
reported their health to be fair or poor, only 12% reported their quality of life to be poor.
Patients’ estimates of their life expectancy were considerably more optimistic than their
physicians’ estimates. While 10% of patients believed they had ≤1 year to live, 41% of
physicians estimated their patients’ life expectancy to be ≤1 year. Of the patients whose
physicians made this estimate, the largest proportion had a diagnosis of cancer (76%), as
compared to COPD (22%) or HF (19%). Only 14% cited hospice as an alternative to
hospitalization, and of these patients, 52% had physicians who reported discussing
hospice, suggesting that few patients knew about hospice outside of a conversation with
their physician.
40
Table 1. Characteristics of older persons with advanced illness
Total
Cancer
Characteristic
(N=215)
(n=79)
Age, %
60-69
35
42
70-79
45
46
80+
20
13
COPD*
(n=79)
HF*
(n=57)
37
43
20
25
47
28
Education ≤ 12 years, %
68
63
75
67
Female, %
42
43
49
30
White†, %
91
92
92
86
Lives alone, %
25
16
27
33
Health perception fair to poor‡, %
71
68
71
75
Quality of life poor to worst possible§, %
12
13
13
11
Selected moderate to severe symptoms, %
Pain‖
29
38
19
29
Decreased activity level§
63
56
73
60
Depression§
13
11
15
13
Shortness of breath‖
42
20
71
32
Unwilling to undergo therapies for return to
current health, %
Major therapies‡
11
9
14
9
Minor therapies§
2
3
1
4
Physicians’ estimate of patient life
expectancy ≤ 1 year, %
41
76
22
19
Physician reported informing patient of life
expectancy, %
30
62
10
14
13
7
Patients’ self-perceived life expectancy ≤ 1
10
10
year¶, %
*COPD = Chronic obstructive pulmonary disease; HF = heart failure
†
N=214; ‡N=211; §N=212; ‖ N=213; ¶N=210
41
Hospice Discussion and Associated Factors
Overall, physicians reported discussing hospice with 22% of patients. The
reasons most frequently cited for not discussing hospice were “not terminally ill” (50%)
and “prognosis too uncertain” (37%). Patient-centered reasons were less frequently cited,
such as “would take away patient’s hope” (10%) and “patient wants life-sustaining
therapies” (9%) (Table 2).
Table 2. Physician reports of hospice discussion and reasons for not discussing hospice
Characteristic
Total
Cancer COPD*
HF*
(N=215) (n=79) (n=79) (n=57)
Physician discussed hospice with patient
22
46
10
7
or family†, %
Reasons for not discussing hospice†, %
Not terminally ill
50
32
66
Prognosis too uncertain
37
29
48
Patient would not handle this discussion
5
0
10
well
Patient wants life-sustaining therapies
9
15
6
Would take away patient’s hope
10
8
19
Services would not benefit patient
9
11
5
*COPD = Chronic obstructive pulmonary disease; HF = heart failure
†
N=214
55
34
4
5
2
13
42
Selected patient- and physician-related characteristics and their association with
hospice discussion are shown in Table 3. Physicians of patients with cancer were more
likely to report a discussion (46%) than physicians of patients with COPD (10%) or with
HF (7%) (P < .001). Among physicians who estimated their patients’ life expectancy to
be ≤1 year, 49% reported discussing hospice, compared to only 4% when they estimated
a longer life expectancy (P < .001). Within the subset of physicians who estimated their
patients’ life expectancy to be ≤1 year, hospice discussion was reported much more
frequently by physicians who were >90% certain about their estimate than by physicians
who were less certain (93% versus 40%, P < .001). Physicians were more likely to report
discussing hospice for patients who self-reported poorer quality of life, moderate to
severe pain, a perceived life expectancy of ≤1 year, an unwillingness to undergo minor
therapies for a return to current health, and that their physician informed them of their life
expectancy. However, 40%-69% of physicians whose patients had these characteristics
did not report a hospice discussion. There was no association between hospice discussion
and patients’ self-rated health or unwillingness to undergo major therapies for a return to
current health.
43
Table 3. Association of patient- and physician-related factors with hospice discussion.
Discussion of
Hospice (n=48)
No Discussion of
Hospice (n=167)
Diagnosis, %
Cancer
Chronic Obstructive Pulmonary Disease
Heart Failure
46
10
7
54
90
93
<.001
Pain*, %
Moderate to severe
None to mild
31
19
69
81
.057
Activity level reduction†, %
Moderate to severe
None to mild
26
17
74
83
.113
Quality of life†, %
Poor or worst possible
Best possible, good, or fair
42
20
58
80
.011
Patient unwilling to undergo minor therapies for
return to current health†, %
Yes
No
60
22
40
78
.043
Patients’ self-perceived life expectancy‡, %
≤1 year
>1 year
41
20
59
80
.028
Physicians’ estimate of patient life expectancy,
%
≤1 year
>1 year
49
4
51
96
<.001
Physicians’ level of certainty about patient life
expectancy when estimate is ≤ 1 year, %
>90%
≤90%
93
40
7
60
<.001
57
7
43
93
<.001
Characteristic (N=215)
Physician reported informing patient of life
expectancy, %
Yes
No
*N=213; †N=212; ‡N=210
P-value
44
In multivariable analysis, physicians’ estimate of patient life expectancy ≤1 year
was the strongest determinant of hospice discussion (odds ratio (OR) = 13, 95%
confidence interval (CI) = 4.3–39) (Table 4). A cancer diagnosis was independently
associated with hospice discussion; other factors associated with hospice discussion in
bivariate analysis did not retain their significance. One variable, “physician informed
patient of life expectancy,” was not entered into the model because of its high correlation
with “physician-estimated life expectancy ≤1 year” (Pearson correlation coefficient,
>0.3). Another variable, “physician certainty about life expectancy,” was not entered into
the model because it was measured only among the sub-group of physicians who
estimated their patients’ life expectancy to be ≤1 year.
Table 4. Multivariable model for characteristics associated with hospice discussion
Variable
Cancer diagnosis
Odds ratio (95%
confidence interval)
3.4 (1.3-8.9)
Male
1.9 (0.8-4.5)
Moderate to severe pain
1.0 (0.4-2.5)
Moderate to severe reduction in
activity level
1.6 (0.7-4.0)
Fair/poor self-rated quality of life
2.1 (0.6-7.6)
Patient self-perceived life
expectancy ≤ 1 year
1.7 (0.5-5.8)
Physicians’ estimate of patient life
expectancy ≤ 1 year
13 (4.3-39)
Patient unwilling to undergo minor
therapies for return to current health
5.2 (0.2-131)
45
Accuracy of Physician Prognosis
A total of 56% of the patients died during the course of the study, including 77%
of patients with cancer, 42% of patients with COPD, and 47% of patients with HF. Of
patients who died, 40% had physicians who estimated that their life expectancy was >1
year, within six months before patient death. Stratified according to patient diagnosis,
physician overestimate of prognosis applied to 11% of patients with cancer and 68% of
patients with COPD or HF.
DISCUSSION
In this study consisting of older adults with advanced cancer, COPD, and HF,
physicians reported discussing hospice for nearly one-half of patients with cancer but
only a small fraction of patients with COPD or HF. Several characteristics suggesting
that patients might benefit from hospice were associated with a greater likelihood of
discussion, including moderate to severe symptoms, unwillingness to undergo minor
medical interventions, and poorer quality of life, but nonetheless, a considerable number
of patients with these characteristics did not have the discussion. Other characteristics of
a similar nature, such as poorer self-rated health and unwillingness to undergo major
medical interventions, were not associated with hospice discussion. The strongest
determinant of hospice discussion was physicians’ estimate of and level of certainty
about patient life expectancy. Nonetheless, physicians were unable to identify as having
a poor prognosis a considerable percentage of patients who subsequently died within six
months.
46
These findings are consistent with a prior study in which physicians reported that
the difficulty of prognostication was the foremost barrier to the physician offering
hospice.54 However, in contrast to the prior study, physicians in this study did not cite
patient readiness to handle the discussion and preferences for treatment as major barriers
to discussion. One conceivable explanation for this discrepancy is that in the current
study physicians were asked about specific patients at specific times rather than general
barriers to discussion. Nonetheless, given evidence from previous research suggesting a
substantial lack of communication between physicians and patients about end-of-life
preferences,20 it is possible that most physicians do not assess whether patients are ready
to handle these discussions or willing to undergo major or minor therapies. Given the
results of this study, it would seem that patients are often not informed about alternatives
to standard therapy because physicians’ discussion of hospice is determined largely by
their assessment of and level of certainty about patient life expectancy.
The close relationship shown in this study between level of prognostic certainty
and hospice discussion appears to imply that more accurate prognostication would
enhance physician-patient communication at the end of life. However, substantial
research evidence indicates the limited value of a quantitative approach to
prognostication for patients approaching the end of life. First, clinical prediction criteria
based on National Hospice Organization guidelines for patients with COPD and HF have
been demonstrated ineffective in recognizing patients with a life expectancy of six
months or less.55 Second, as previously discussed, providing state-of-the-art prognostic
information to physicians in the SUPPORT study did not improve physician-patient
communication.20
47
Although communication and care at the end of life will be unlikely to improve
with greater prognostic accuracy, nonetheless, the results of this study confirm that they
may be improved by adequately addressing the problem of prognostication. Given the
relatively narrow approach to prognosis taken in the past century, the practice of
prognostication is unlikely to improve significantly without a dramatically new approach
that is based on sound science but less statistically oriented. One example of this, as
stated above, is the suggestion for altering the Medicare Hospice Benefit requirements to
include criteria such as functional status, symptom burden, and quality of life.30 This
alternative approach would potentially include the many patients in this study whose state
of health or preferences suggest that they may have benefited from hospice services, but
who did not benefit from discussions with their physicians about hospice. However,
more research is needed to further support this approach, which may not gain easy
acceptance among researchers or policymakers who have primarily focused for decades
on a quantitative approach.
One of the limitations of this study is that descriptions of the nature of discussions
about hospice were not obtained, including whether physicians simply provided
information about services or also made a recommendation; nonetheless, given the large
proportion of patients in this study who utilized hospice following the discussion, it
would seem that discussions were mainly characterized by the latter. A second limitation
is that information was obtained by self-report, with no confirmation as to whether the
discussions reported by physicians took place. Physicians, who completed surveys at sixmonth intervals, may have been asked to recall discussions that took place months prior.
48
Additionally, desirability bias may have influenced the responses. It is also possible that
physicians were more likely to have discussions as a result of participating in the study.
In this study, physicians’ decisions to discuss hospice for older persons with
advanced illness were influenced mainly by an approach to prognostication that focuses
on estimates of life expectancy and the predictability of disease course. It is important
for physicians to have such discussions with terminally ill patients so that patients can
understand their options and make informed decisions about their care at the end of life.
Since prognostication for patients with non-cancer diagnoses has particular limitations,
hospice discussions occur primarily for patients with cancer near the end of life. Many
persons who might benefit from hospice, as suggested by their health status and treatment
preferences, are not having these discussions with their physicians. Based on prior
evidence, more accurate prognostication tools are unlikely to improve communication
between physicians and patients at the end of life, thus necessitating a different approach
to prognostication that is less numerical and more conceptual in orientation.
The next section will further explore the possibilities for a more conceptual
approach to prognostication for elderly persons nearing the end of life, by means of a
thorough literature review and subsequent synthesis of data.
49
PART II: Systematic Review
Despite a recent increased interest in prognostication, the practice of it remains
rare because of a limited conceptual understanding and approach. As a primary example,
prognostic models, which are popular in clinical research but not clinical practice,
erroneously strive to achieve a level of certainty about life expectancy that can be applied
to individuals. The technical problems with prognostic models are numerous: their
overall accuracy is only modest,25,55, they have not been externally validated in most
cases,26 and they contain varying combinations of specific factors that may not be readily
measured in all clinical settings. Furthermore, physicians have reported that they prefer
to avoid specific estimates when discussing prognosis with patients,56,57 which is
consistent with evidence from the SUPPORT trial, in which the availability of more
accurate prognostic information did not improve communication between physicians and
patients.20 Finally, although physicians generally agree that discussions about end-of-life
issues occur too late,40,58 there is no consensus about what clinical markers might prompt
discussions58,59 and prognostic models of mortality do not provide this information.
Alternatively, a broader understanding of the domains of health most strongly
associated with mortality might allow for assessments aimed simply at identifying
patients with increased mortality risk, so that a deficit in one of the most important
domains would prompt re-evaluation of the approach to care. This alternative may be
more acceptable to physicians than calculations of absolute mortality risk, and may be
more readily used in time-limited clinical settings. It also serves to broaden the approach
to prognostication, which has been severely restricted in recent decades to a quantitative
estimate of life expectancy.
50
A systematic review has the potential to take a comprehensive compilation of all
known factors associated with mortality and to identify larger patterns of association
across studies. This information may then be used to identify broader categories of
health-related characteristics associated with mortality. No such review now exists in the
literature, the closest being one review60 that examined factors associated with a number
of outcomes in older hospitalized patients, but their goal was to provide a system for
measuring hospital case mix. They evaluated length of stay, discharge destination, and
readmission rates in addition to mortality, and they combined in-hospital mortality and
mortality up to 2 years following admission as outcomes in their analyses.
The goals of this systematic review were as follows: 1) to identify health-related
characteristics of older hospitalized patients and nursing home residents associated with
short-term mortality (1 year or less), 2) to classify these characteristics into domains of
health, and 3) to determine the relative strength of association of these domains of health
with mortality.
51
METHODS
Data Sources and Searches
Mr. Thomas performed an electronic literature search of all English articles
published in MEDLINE (1948-), Scopus (1960-), or Web of Science (1899-) before
August 1, 2010 to identify prospective cohort studies on factors associated with shortterm mortality in elderly hospitalized patients and nursing home residents. The
MEDLINE search used a combination of “filters,” consisting of MeSH terms,
subheadings, text words, and multi-purpose terms, designed to maximize sensitivity and
specificity for this topic. The following filters were used: prognosis studies, mortality,
predictors, age, hospitalized patients, and nursing home residents. Modified forms of the
same filters were used for the Scopus and Web of Science searches (Table 5).
52
Table 5. Literature search strategies for MEDLINE, Scopus, and Web of Science
MEDLINE Search Strategy
Prognosis studies filter
1. incidence/
2. exp mortality/
3. Follow-Up Studies/
4. mortality.fs.
5. prognos:.tw.
6. predict:.tw.
7. course.tw.
8. outcome:.tw.
9. 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8
Predictors filter
1. risk:.mp.
2. assess:.mp.
3. predict:.mp.
4. factor:.mp.
5. screen:.mp.
6. probability:.mp.
7. exp risk/
8. 1 or 2 or 3 or 4 or 5 or 6 or 7
Mortality filter
1. exp mortality/
2. exp death/
3. exp survival analysis/
4. Life Expectancy/
5. mortality.fs.
6. death.mp.
7. survival.mp.
8. mortality.mp.
9. die:.mp.
10. 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or
9
Age filter
1. exp aged/
Prognosis study filter and Mortality filter and Predictors filter and Age filter =
Combo filter
Combo filter was then combined with the hospitalized patients filter and with the nursing
home residents filter in separate searches.
Hospitalized patients filter
1. hospital:.ti. and (elder: or old: or geriatric:).mp
2. (elder: adj2 hospitali#ed).mp
3. (old: adj2 hospitali#ed).mp
4. (geriatric: adj2 ward:).mp.
5. (geriatric: adj2 unit:).mp.
6. intensive care.ti. and (elder: or old: or geriatric:).mp.
7. inpatient:.ti. and (elder: or old: or geriatric:).mp.
8. geriatric: hospital:).mp.
9. ICU.ti. and (elder: or old: or geriatric:).mp.
10. intermediate care.ti. and (elder: or old: or geriatric:).mp.
11. (ward or wards).ti. and (elder: or old: or geriatric:).mp.
12. ((acute: adj2 hospital:) and (elder: or old: or geriatric:)).mp.
13. 1 or 2 or 3 or 4 or 5 or 6 or 7 or 8 or 9 or 10 or 11 or 12
53
Table 5 (continued). Literature search strategies for MEDLINE, Scopus, and Web
of Science
Nursing home residents filter
1. exp Residential Facilities/
2. Long-Term Care/
3. Institutionalization/
4. nursing home:.ti.
5. long-term care.ti.
6. extended care.ti.
7. 1 or 2 or 3 or 4 or 5 or 6
Scopus and Web of Science Search Strategy
Prognosis studies filter
mortality OR “follow up” OR outcome
OR outcomes OR prognosis OR
prognoses
OR predict OR predicts OR predictor
OR predictors
OR survival OR “survival analysis” OR
“survival analyses” OR die*
Predictors filter
risk OR risks OR screen* OR factor OR
factors OR predict OR predicts OR
predictor OR predictors
Mortality filter
mortality OR death OR “life
expectancy”
Age filter
elder OR elders OR elderly OR old OR
older OR geriatric OR geriatrics
Prognosis study filter and Mortality filter and Predictors filter and Age filter =
Combo filter
Combo filter was then combined with the hospitalized patients filter and with the nursing
home residents filter in separate searches.
Nursing home residents filter
“Nursing home” OR “nursing homes” OR “long term care” OR “extended care” OR
“assisted living” OR Institutionali*
Hospitalized patients filter
Title (hospital* OR ward OR wards OR “intensive care” OR ICU)
54
Study Selection
A total of 7,644 articles were identified. Of these, 5,393 were excluded by Mr.
Thomas through title review, and the remaining 2,251 were reviewed in abstract form by
Mr. Thomas and Dr. Cooney independently. The 367 articles appearing to meet inclusion
criteria, as presented below, based on review of the abstract alone were then retrieved and
examined in full text. Mr. Thomas and Dr. Fried reviewed the articles to determine
whether they met inclusion criteria, and consensus was achieved, resulting in a total of 45
articles included in this study. To ensure completeness, the reference lists of the 45
included articles were also reviewed by Mr. Thomas (by title at least, and if necessary, by
abstract and full text), resulting in three additional articles. In the same way, all articles
published in Journal of the American Geriatrics Society from 2001 to 2010 were
reviewed by Mr. Thomas, but no additional articles fulfilling inclusion criteria were
identified (Figure 1).
55
Figure 1. Summary of literature search and selection.
56
The final inclusion list contained 48 articles. These articles were separated into
the following categories based on patient population and follow-up period: general
nursing home residents, nursing home residents with advanced dementia, hospitalized
patients with in-hospital mortality, and hospitalized patients with mortality up to one year
following admission.
Inclusion Criteria
Studies were included if their participants were hospitalized patients, long-stay
nursing home residents, or nursing home residents with advanced dementia only,
regardless of length of stay. Long-stay was defined as residing in the nursing home for at
least 3 months, and was specified to avoid cohorts containing patients receiving shortterm rehabilitation or other forms of subacute care. We only included studies of persons
age 65 years or older, or, if age range was not provided, studies with a population average
age of ≥80 years. We only included studies with a prospective cohort design, including
studies that use chart or record review in a prospective fashion; studies that examine at
least one health-related characteristic, meaning a characteristic inherent to the patient and
not involving elements of health care treatment, medical devices, or living location;
studies that measure mortality within a follow-up period of one year or less; and studies
that contain at least bivariate analysis. The criterion of a follow-up period of one year or
less was chosen based on a cursory review of the literature in which a considerable
number of studies on mortality specified a follow-up period of one year.
We did not include studies that excluded patients possessing characteristics
potentially associated with mortality. Specifically, we did not include studies of nursing
home residents that excluded terminally ill persons, those receiving palliative care, or
57
those with specific illnesses, medications, or nutritional requirements; and we did not
include studies of hospitalized patients that excluded nursing home residents, the
terminally ill, patients receiving palliative care, those with specific illnesses or
abnormalities, those who died in the hospital, or those living outside a prescribed
geographical area.
Data Extraction and Quality Assessment
We first sought to identify all markers for short-term mortality, regardless of
whether they have independent association with mortality, because of their potential
clinical usefulness even if they are subject to confounding. Thus, from all studies
fulfilling inclusion criteria, Mr. Thomas generated a list of health-related characteristics
associated with short-term mortality in bivariate analysis (p<0.05) and organized them
according to the categories of patient populations described above.
For the purpose of quality assessment, we modeled a set of six criteria after
recommendations specific to prognosis studies in systematic reviews61 and applied them
to the 48 articles meeting the inclusion criteria. Twenty-four articles were determined to
be of high quality, in that they fulfill all our criteria (Table 6).
58
Table 6. Criteria used for assessing quality of articles that met inclusion criteria.
Study population: For patients who could not be interviewed, we required that the selfreported data be sought from proxies.
Attrition: We required that loss to follow-up be no greater than 20%.
Prognostic factors: The reliability of data collection was considered for the following
variables: dementia, delirium, malnutrition or malnourishment, and depression. For these
variables, clinical assessment rather than medical chart review was required.
Confounders: Because a determination of independent association was not required for
this review, we did not include criteria related to confounders.
Statistical analysis: We required that the number of outcomes be at least ten times the
number of variables in the model.
Study design: For studies that utilized secondary analysis of previously collected data,
we assessed whether the original study was an observational cohort study or a
randomized controlled trial. Because randomized controlled trials may not include
representative populations,62 we considered only studies derived from observational
cohort studies to fulfill the quality criterion.
59
Data Synthesis and Analysis
In order to organize the health-related characteristics associated with mortality
and allow for broader, more meaningful comparisons, all three investigators (Mr.
Thomas, Dr. Fried, and Dr. Cooney) reached consensus in grouping these characteristics
into larger aspects of patient health, or domains. Demographic information was not
placed in a domain or included in subsequent analyses because the purpose of the review
was to focus on health-related factors.
The heterogeneity of study populations, independent variables measured, and
statistical methods precluded combining results in a meta-analysis. Therefore, Mr.
Thomas and Dr. Fried developed methods of summarizing the strength of association of
the different characteristics and domains with mortality across studies, based on the
following: 1) the frequency, across individual studies, with which particular healthrelated characteristics and domains were associated with mortality in bivariate and
multivariable analysis; and 2) the relative ranking, within individual studies, of the
strength of association with mortality for each domain of health-related characteristics.
Some single studies contained more than one cohort, and some pairs of studies
contained identical cohorts. For the data extraction step (performed by Mr. Thomas), we
included such instances of repetition to ensure completeness in identifying all
characteristics associated with mortality. In the analysis steps (performed by Mr.
Thomas), we excluded repetitive data as follows: when studies consisted of both
development and validation cohorts, data from the validation cohort was preferentially
used, unless analyses were available only for the development cohort; and when studies
were found to contain identical or overlapping populations, a single study was chosen
60
from among them on a case-by-case basis (Table 8).
Individual characteristics were included in the first step of analysis if they were
measured in more than one article category or in more than one study within a given
category. For every article that evaluated a given characteristic, we noted whether an
association was found in bivariate analysis, and if assessed, in multivariable analysis.
Then we calculated frequencies by dividing the number of times a characteristic was
associated with mortality by the total number of times assessed for both bivariate and
multivariable analysis. Finally, we combined the frequencies for all characteristics
within each domain to produce an overall calculation of the frequency with which a
domain was associated with mortality out of the total number of times assessed in
bivariate and multivariable analysis.
In order to accomplish the second step, which summarized across studies the
relative strength of association of a given domain compared to the other domains, we
developed and executed a method of calculation based on multivariable analysis data. To
achieve this comparison, articles containing multivariable analysis were selected, and
hazard ratios or odds ratios from these multivariable models were used to rank
statistically significant (p<0.05) health-related characteristics from highest to lowest
association within each study. Health-related characteristics were then linked with their
respective domains, and a ranking of the relative importance of domains for each article
was generated. If more than one factor was associated with a single domain, the domain
was assigned to the strongest factor. Finally, domains that were investigated but did not
attain significance in multivariable analysis were placed at the bottom of the ranking list
and set equal to each other.
61
The ranking of domains within each study was then compared across studies to
generate an overall ranking of the relative importance of each domain. Specifically, the
number of instances a given domain outranked any other domain was divided by the total
number of instances in which that domain was ranked against any other domain; this
frequency was then multiplied by 100% to achieve a head-to-head ranking percentage
(see Table 7 for an example). These percentages, which we will refer to as measures of
“relative strength,” were then compared to assign importance to domains. This analysis
was also performed for the subset of articles assessed to be of high quality.
Table 7. Example of head-to-head domain analysis to determine the relative
strength of association with mortality for each domain compared to other domains.
Nursing home residents (all articles regardless of quality)
cognitive
function
disease
diagnosis
physical
function
nutrition
pressure
sores
shortness
of breath
head-tohead totals
0:3
0:3
0:2
0:0
0:2
0/10 = 0%
2:1
0:2
2:0
0:2
7/12 = 58%
1:1
2:0
1:1
8/12 = 75%
2:0
1:1
8/10 = 80%
cognitive
function
disease
diagnosis
physical
function
3:0
3:0
1:2
nutrition
2:0
2:0
1:1
pressure
0:0
0:2
0:2
0:2
0:2
0/8 = 0%
sores
shortness
2:0
2:0
1:1
1:1
2:0
8/10 = 80%
of breath
Each ratio in the white boxes above indicates a head-to-head comparison between two domains. To achieve
an overall head-to-head ranking percentage for each domain, the number of instances a given domain
outranked any other domain was divided by the total number of instances in which that domain was ranked
against any other domain; this frequency was then multiplied by 100%.
62
Since not all domains were examined in all articles containing multivariable
analysis, there was greater evidence supporting the “relative strength” of some domains
as compared to others. To our knowledge, no criteria exist for evaluating this variation in
the amount of evidence for domains. Hence, Mr. Thomas and Dr. Fried devised criteria
that consider both quality and quantity of evidence, and made note of each domain that
did not fulfill at least one of the following in its respective article category: examined in
at least 2 high quality articles, examined in at least 1 high quality article with a
participant population of 10,000 or more, or examined in at least 4 articles regardless of
quality.
RESULTS
Our literature search identified 48 articles63-110 published before August 2010 that
met the inclusion criteria (Table 8). Nine studies involved general nursing home
residents, two studies involved nursing home residents with advanced dementia, 24
studies involved hospitalized patients with in-hospital mortality, and 17 studies involved
hospitalized patients with mortality up to one year.
63
Table 8. Studies on health-related characteristics associated with short-term
mortality in the elderly
General nursing home residents
Reference
Year
Country
Sample size
Follow-up
Quality
Domains
63
Barca et al.
2010 Norway
902
1 year
+
3
Flacker et al.64*
1998 USA
780
1 year
5
65
Flacker et al. †
2003 USA
15,068
1 year
+
5
Grabowski et al.66
2005 USA
2,782
1 year
+
1
Kiely et al.67*
2000 USA
778
1 year
5
Kiely et al.68†
2002 USA
33,188
1 year
+
5
69
Mooradian et al.
1991 USA
129
4 months
+
1
Perls et al.70
1993 USA
1,951
6 months
+
2
Tsai et al.71
2008 Taiwan
308
1 year
+
1
*†These pairs of studies used the same population for their respective analyses; the Kiely articles separated
analyses by gender and were chosen for exclusion from analysis in this review.
Nursing home residents with advanced dementia
Reference
Year
Country
Sample size
72
Mitchell et al.
2004 USA
4,631
Mitchell et al.73
2010 USA
222,405
Follow-up
6 months
1 year
Quality
+
+
Domains
4
5
64
Table 8 (continued). Studies on health-related characteristics associated with shortterm mortality in the elderly
Hospitalized patients: in-hospital mortality
Reference
Year
Country
Sample size
Follow-up
Quality
Domains
91
Abizanda et al.
2007 Spain
356
in-hospital
+
1
Agarwal et al.92
1988 USA
80
in-hospital
3
Alarcon et al.74
1999 Spain
353
in-hospital
2
Bienia et al.93
1982 USA
59
in-hospital
+
1
77
Covinsky et al.
1997 USA
823
in-hospital
2
Eeles et al.80
2010 UK
278
in-hospital
1
Gazzotti et al.94
2000 Belgium
175
in-hospital
+
5
Incalzi et al.95
1992 Italy
308
in-hospital
+
4
Incalzi et al.96*
1996 Italy
302
in-hospital
4
97
Incalzi et al. *
1997 Italy
370
in-hospital
5
Iwata et al.98
2006 Japan
1638
in-hospital
+
2
Jonsson et al.99
2008 Iceland
749
in-hospital
3
Marengoni et al.100† 2003 Italy
923
in-hospital
5
Marengoni et al.101† 2008 Italy
596
in-hospital
5
85
Narain et al.
1988 USA
396
in-hospital
+
2
O'Keeffe et al.102
1997 Ireland
225
in-hospital
+
1
Pompei et al.103
1994 USA
323
in-hospital
1
Ponzetto et al.104
2003 Italy
987
in-hospital
5
Sampson et al.105
2009 UK
617
in-hospital
+
3
106
Sonnenblick et al.
2007 Israel
779
in-hospital
+
4
Stratton et al.107
2006 UK
150
in-hospital
+
1
Terzian et al.108
1994 USA
4,123
in-hospital
+
2
Zafrir et al.109
2010 Israel
333
in-hospital
+
6
Zekry et al.110
2010 Switzerland
444
in-hospital
1
*†These pairs of studies appear to have used the same study population for their respective analyses; Incalzi
et al. 1997 was chosen for analysis because it evaluated a greater number of domains, and Marengoni et al.
2003 was chosen because it more thoroughly evaluated physical functional measures.
65
Table 8 (continued). Studies on health-related characteristics associated with shortterm mortality in the elderly
Hospitalized patients:
Reference
Alarcon et al.74
Boyd et al.75*
Buurman et al.76
Covinsky et al.77
Desai et al.78†
Drame et al.79
Eeles et al.80
Flodin et al.81
Gonzalez et al.82
Inouye et al.83
Laurila et al.84‡
Narain et al.85
Persson et al.86
Pilotto et al.87
Pitkala et al.88‡
Van Doorn et al.89†
Walter et al.90*
mortality up to one year
Year
Country
Sample size
1999 Spain
353
2008 USA
2,279
2008 Netherlands
463
1997 USA
823
2002 USA
524
2008 France
1,306
2010 UK
278
2000 Sweden
552
2009 Chile
542
2003 USA
1,246
2004 Finland
425
1988 USA
396
2002 Sweden
83
2008 Italy
857
2005 Finland
425
2001 USA
524
2001 USA
1495
Follow-up
6 months
1 year
3 months
1 year
1 year
6 weeks
1 year
1 year
3 months
1 year
1 year
6 months
1 year
1 year
1 year
1 year
1 year
Quality
+
+
+
+
Domains
3
1
5
2
4
5
1
3
3
5
1
4
1
5
1
1
5
*†‡Duplicate study populations were used for these pairs of studies in their respective analyses; Walter et al.
and Desai et al. were selected for analysis because they evaluated a greater number of domains, and Pitkala et
al. was selected because it included multivariable analysis.
66
All health-related characteristics associated with short-term mortality in one or
more studies are listed in Table 9. We classified the characteristics into seven domains:
cognitive function, disease diagnosis, physical function, laboratory values, nutrition,
pressure sores, and shortness of breath. Characteristics with ambiguous classification
were placed into domains on a case-by-case basis (e.g., albumin was placed under
nutrition, and delirium under cognitive function). Since laboratory values were available
in studies involving hospitalized patients but not in those involving nursing home
residents, and reports of shortness of breath were only available in the latter, each article
category contained six domains rather than seven.
67
Table 9. Patient health-related characteristics associated with short-term mortality
in at least one study
General nursing home residents
Cognitive function
Ability to understand or be
understood
Change in cognitive status
CPS
Decision-making ability
Decline in cognitive function
Delirium/delirium symptoms
Dementia
Long-term memory impairment
Short-term memory impairment
Disease diagnosis
Acute episode
Anemia
Asthma/emphysema/COPD
Cancer
Congestive heart failure
Cardiovascular diseases
Depression
Disease diagnosis (cont’d)
Detectable TNF levels
Diabetes mellitus
Fever
Physical health rating
Peripheral vascular disease
Renal disease/failure
Stroke
Unstable condition
Urinary tract infection
Physical function
ADL score
Balance problem
Bed rail use
Bedfast all or most of time
Bowel incontinence
Decline in ADLs
Hearing problem
Not awake most of day
Physical function (cont’d)
Recent fall
Vision problem
Nutrition
>25% of food uneaten
BMI
Chewing problem
Mechanically altered diet
MNA
Refuses fluids
Swallowing problems
Weight loss
Pressure sores
Shortness of breath
Nursing home residents with advanced dementia
Cognitive function
Absence of Alzheimer's disease
Cognitive deterioration
CPS
Hallucinations or delusions
Rarely understood
Disease diagnosis
Anemia
Asthma/emphysema/COPD
Cancer
Cardiac Dysrhythmia
Congestive heart failure
Diabetes mellitus
Edema
Fever
Heart disease
Disease diagnosis (cont’d)
Hip fracture
Hypertension
Infection
No seizure disorder
Non-hip fracture
Oxygen therapy
Parkinson's disease
Peripheral vascular disease
Pneumonia or RTI
Renal failure
Septicemia
Stroke
Unstable medical condition
Urinary tract infection
Physical function (cont’d)
Aspirations
Bedfast
Bowel incontinence
Functional deterioration
Not awake most of day
Physical function
ADL score
Shortness of breath
Nutrition
<25% of food eaten
BMI
Chewing or swallowing problem
Insufficient fluid intake
Weight loss
Pressure sores
ADLs=Activities of Daily Living; BMI=Body Mass Index; COPD=Chronic Obstructive Pulmonary Disease;
CPS=Cognitive Performance Scale; MNA=Mini-Nutritional Assessment; RTI=Respiratory Tract Infection.
68
Table 9 (continued). Patient health-related characteristics associated with shortterm mortality in at least one study
Hospitalized patients with in-hospital mortality
Cognitive function
Cognitive function score
CPS
Delirium (DSM-IV, unspecified)
Dementia (DSM-IV, Carmel
Hospital scale)
Level of confusion/
consciousness
MSQ
MMSE
Moderate to Severe Dementia
Short Portable MSQ
Disease diagnosis
4 or more diagnoses
APS
Admission for new problem &
exacerbated old problem
Age-comorbidity index
APACHE II score
APACHE score
Atrial fibrillation
Cancer
CCI
Cerebrovascular disease
CIRS co-morbidity
Disease diagnosis (cont’d)
Gastrohepatic disease
GDS
Heart diseases
Index of Coexisting Disease
Index of comorbidity
Infectious disease
Inotropic therapy
Kaplan score
Mechanical ventilation
No operation
Number of additional diagnoses
One or more procedures
Pseudomembranous colitis
Sepsis
Physical function
ADL dependency
ADLs on admission
Barthel index on admission
Instrumental ADLs
Katz scores
Preadmission ADL impairment
Red Cross Hospital FDS
Upper extremity function
Lab values
Cholesterol
Creatinine
Elevated CRP
Fibrinogen
HDL cholesterol
Hemoglobin
Hyponatremia
Leukocyte count
Neutrophil count
Total lymphocyte count
Transferrin
Urea
Nutrition
Albumin <3.5, <3, and <2.8
BMI
Prognostic nutritional index
MUST
Pressure sores
Other
Diastolic blood pressure
Heart rate
Systolic blood pressure
69
Table 9 (continued). Patient health-related characteristics associated with short-term
mortality in at least one study
Hospitalized patients with mortality up to one year
Cognitive function
Disease diagnosis (cont’d)
Delirium (DSM-IV, ICD-10) or
Malignancy
duration of delirium
Metastatic cancer
Dementia
Moderate/severe renal disease
MSQ
Pneumonia
Short Portable MSQ
Respiratory failure
Severe PVD
Disease diagnosis
Solitary cancer
Acute or chronic renal failure
APS
Physical function
Bone marrow failure
ADLs on admission, during stay,
Cerebrovascular disease
or at discharge
CCI
Barthel index on admission or
CIRS co-morbidity
during stay
Congestive heart failure
Instrumental ADLs
COPD
Katz scores
Depression
New self-care ADL disability
History of myocardial infarction
Pfeffer functional score
High risk diagnosis group
Urinary incontinence
Lymphoma/leukemia
Walking impairment
Lab values
BUN
Creatinine
Hematocrit
Nutrition
Albumin <3.5 and <4
BMI
Malnutrition
MNA
MNA Short Form
SGA
Pressure sores
Presence of pressure sores
Pressure sore risk (Norton scale,
Exton-Smith scale)
Other
MP
ADLs=Activities of Daily Living; APS=Acute Physiology Score; BMI=Body Mass Index; BUN=Blood Urea Nitrogen;
CCI=Charlson Comorbidity Index; CIRS=Cumulative Illness Rating Scale ; CPS=Cognitive Performance Scale;
CRP=C-Reactive Protein; FDS=Functional Disability Scale; GDS=Geriatric Depression Scale; MMSE=Mini-Mental
State Examination; MNA=Mini Nutritional Assessment; MPI=Multidimensional Prognostic Index; MSQ=Mental
Status Questionnaire; MUST=Malnutrition Universal Screening Tool; PVD=Peripheral vascular disease;
SGA=Subjective Global Assessment.
70
Frequency of association of characteristics with mortality across studies in bivariate
and multivariable analyses
The 48 studies that met inclusion criteria reported on data from 41 unique
populations, so seven studies were excluded from the analysis steps (Table 8). Healthrelated characteristics selected according to the criteria described in the methods section,
and their association with mortality in bivariate and multivariable analysis, are shown in
Table 10. Many characteristics that were associated with mortality in bivariate analysis
failed to retain significance in multivariable analysis.
General nursing home residents
Poorer physical function, poorer nutrition, and shortness of breath were
significant in 100% of bivariate and multivariable analyses. Poorer cognitive function
and the presence of pressure sores were associated with mortality in 100% of bivariate
analyses but were not significant in any multivariable analyses. Disease diagnosis was
associated with mortality in 89% of bivariate analyses and in 56% of multivariable
analyses.
Nursing home residents with advanced dementia
All health-related characteristics were associated with mortality in all bivariate
analyses. Shortness of breath was significantly associated with mortality in both
multivariable analyses. Poorer physical function, pressure sores, and poorer nutrition
were significantly associated with mortality in 64%, 50%, and 44% of multivariable
analyses, respectively. Disease diagnosis was significantly associated with mortality in
only 15% of multivariable analyses.
71
Hospitalized patients: in-hospital mortality
Pressure sores were significantly associated with mortality in the only study that
examined it in both bivariate and multivariable analysis. Poorer cognitive function,
poorer physical function, and poorer nutrition were significantly associated with
mortality in 100%, 88%, and 88% of bivariate analyses, respectively, and in 64%, 57%,
and 45% of multivariable analyses, respectively. Disease diagnosis and poorer laboratory
values were significantly associated with mortality in 74% and 75% of bivariate analyses,
respectively, and in 36% and 21% of multivariable analyses, respectively.
Hospitalized patients: mortality up to one year following admission
Poorer nutrition and the presence of pressure sores were significantly associated
with mortality in 100% of bivariate analyses and in 86% and 50% of multivariable
analyses, respectively. Poorer physical function was significantly associated with
mortality in 87% of bivariate analyses and 67% of multivariable analyses. Poorer
cognitive function, disease diagnosis, and poorer laboratory values were significantly
associated with mortality in 69%, 73%, and 50% of bivariate analyses, respectively, and
in 54%, 35%, and 50% of multivariable analyses, respectively.
72
Table 10. Selected health-related patient characteristics and their frequency of
association with mortality across studies in bivariate and multivariable analyses
General nursing home residents
Bivariate
Analyses
Multivariable
Analyses
Cognitive function
All cognitive measures
CPS
Delirium/delirium symptoms
Dementia diagnosis
Short-term memory problem
9/9 (100%)
1/1 (100%)
1/1 (100%)
3/3 (100%)
1/1 (100%)
0/9 (0%)
0/1 (0%)
0/1 (0%)
0/3 (0%)
0/1 (0%)
Disease diagnosis
All disease diagnoses
Anemia
Cancer
Congestive heart failure
Diabetes
Renal failure
Unstable condition
17/19 (89%)
2/2 (100%)
2/3 (67%)
2/2 (100%)
2/2 (100%)
0/0 (0%)
2/2 (100%)
9/16 (56%)
0/2 (0%)
1/3 (33%)
2/2 (100%)
1/2 (50%)
0/0 (0%)
1/2 (50%)
Physical function
ADLs
3/3 (100%)
3/3 (100%)
Nutrition
All nutritional measures
>25% food uneaten
BMI
Swallowing problem
Weight loss
8/8 (100%)
1/1 (100%)
3/3 (100%)
1/1 (100%)
2/2 (100%)
6/6 (100%)
1/1 (100%)
2/2 (100%)
1/1 (100%)
2/2 (100%)
Pressure sores
2/2 (100%)
0/2 (0%)
Shortness of breath
2/2 (100%)
2/2 (100%)
ADLs=Activities of Daily Living; BMI=Body Mass Index; CPS=Cognitive Performance Scale
73
Table 10 (continued). Selected health-related patient characteristics and their
frequency of association with mortality across studies in bivariate and multivariable
analyses
Nursing home residents with advanced dementia
Bivariate
Analyses
Cognitive function
All cognitive measures
5/5 (100%)
Absence of Alzheimer's
1/1 (100%)
CPS
1/1 (100%)
Decline in cognitive function
1/1 (100%)
Multi-variable
Analyses
0/5 (0%)
0/1 (0%)
0/1 (0%)
0/1 (0%)
Disease diagnosis
All disease diagnoses
Anemia
Cancer
Congestive heart failure
Dehydration
Diabetes
Renal failure
Unstable condition
33/33 (100%)
1/1 (100%)
2/2 (100%)
2/2 (100%)
1/1 (100%)
2/2 (100%)
1/1 (100%)
1/1 (100%)
5/33 (15%)
0/1 (0%)
1/2 (50%)
2/2 (100%)
0/1 (0%)
0/2 (0%)
0/1 (0%)
1/1 (100%)
Physical function
All functional measures
ADLs
ADL decline
Aspirations
Bowel incontinence
Bedfast
Not awake most of day
11/11 (100%)
2/2 (100%)
1/1 (100%)
2/2 (100%)
2/2 (100%)
2/2 (100%)
2/2 (100%)
7/11 (64%)
2/2 (100%)
0/1 (0%)
0/2 (0%)
2/2 (100%)
2/2 (100%)
1/2 (50%)
Nutrition
All nutritional measures
<25% food eaten
BMI
Swallowing problem
Weight loss
9/9 (100%)
1/1 (100%)
2/2 (100%)
2/2 (100%)
2/2 (100%)
4/9 (44%)
1/1 (100%)
1/2 (50%)
0/2 (0%)
1/2 (50%)
Pressure sores
2/2 (100%)
1/2 (50%)
Shortness of breath
2/2 (100%)
2/2 (100%)
ADLs=Activities of Daily Living; BMI=Body Mass Index; CPS=Cognitive Performance Scale
74
Table 10 (continued). Selected health-related patient characteristics and their
frequency of association with mortality across studies in bivariate and multivariable
analyses
Hospitalized patients: in-hospital mortality
Bivariate
Analyses
MultiVariable
Analyses
Cognitive function
All cognitive measures
CPS
Delirium
Dementia diagnosis
MMSE
MSQ
15/15 (100%)
0/0 (0%)
3/3 (100%)
1/1 (100%)
4/4 (100%)
1/1 (100%)
7/11 (64%)
1/1 (100%)
2/3 (67%)
1/1 (100%)
3/4 (75%)
0/1 (0%)
Disease diagnosis
All disease diagnoses
Cancer
CCI
GIC
35/47 (74%)
2/3 (67%)
4/4 (100%)
2/2 (100%)
17/47 (36%)
2/3 (67%)
0/4 (0%)
2/2 (100%)
Physical function
All functional measures
ADLs
Barthel index
Upper extremity function
14/16 (88%)
6/6 (100%)
2/2 (100%)
1/1 (100%)
8/14 (57%)
4/5 (80%)
1/2 (50%)
1/1 (100%)
Laboratory values
All laboratory values
BUN
Creatinine
Hematocrit
Hemoglobin
Lymphocyte count
Sodium
15/20 (75%)
1/1 (100%)
2/2 (100%)
0/0 (0%)
1/2 (50%)
2/3 (67%)
2/2 (100%)
4/19 (21%)
0/1 (0%)
1/2 (50%)
0/0 (0%)
0/2 (0%)
1/3 (33%)
1/2 (50%)
Nutrition
All nutritional measures
Albumin
BMI
MNA
15/17 (88%)
7/7 (100%)
1/2 (50%)
1/1 (100%)
5/11 (45%)
4/6 (67%)
0/1 (0%)
0/0 (0%)
Pressure sores
1/1 (100%)
1/1 (100%)
ADLs=Activities of Daily Living; BMI=Body Mass Index; BUN=Blood urea nitrogen; CCI=Charlson Comorbidity
Index; CPS=Cognitive Performance Scale; GIC=Geriatrics Index of Comorbidity; MMSE=Mini-Mental State
Examination; MNA=Mini-Nutritional Assessment; MSQ=Mental Status Questionnaire
75
Table 10 (continued). Selected health-related patient characteristics and their
frequency of association with mortality across studies in bivariate and multivariable
analyses
Hospitalized patients: mortality up to one year
Bivariate
Analyses
MultiVariable
Analyses
Cognitive function
All cognitive measures
CPS
Delirium
Dementia diagnosis
MMSE
MSQ
9/13 (69%)
0/0 (0%)
4/6 (67%)
3/4 (75%
0/0 (0%)
1/1 (100%)
7/13 (54%)
0/0 (0%)
4/6 (67%)
1/4 (25%)
0/0 (0%)
1/1 (100%)
Disease diagnosis
All disease diagnoses
Cancer
CCI
Congestive heart failure
GIC
29/40 (73%)
3/3 (100%)
3/3 (100%)
2/2 (100%)
0/0 (0%)
14/40 (35%)
3/3 (100%)
2/3 (67%)
2/2 (100%)
0/0 (0%)
Physical function
All functional measures
ADLs
Barthel index
Katz score
Upper extremity function
13/15 (87%)
4/4 (100%)
3/3 (100%)
2/2 (100%)
0/0 (0%)
10/15 (67%)
4/4 (100%)
2/3 (67%)
2/2 (100%)
0/0 (0%)
Laboratory values
All laboratory values
BUN
Creatinine
Hematocrit
Hemoglobin
Lymphocyte count
Sodium
4/8 (50%)
1/1 (100%)
2/3 (67%)
1/1 (100%)
0/1 (0%)
0/0 (0%)
0/1 (0%)
4/8 (50%)
1/1 (100%)
2/3 (67%)
1/1 (100%)
0/1 (0%)
0/0 (0%)
0/1 (0%)
Nutrition
All nutritional measures
Albumin
BMI
MNA or Short MNA
10/10 (100%)
2/2 (100%)
1/1 (100%)
4/4 (100%)
6/7 (86%)
2/2 (100%)
1/1 (100%)
2/2 (100%)
Pressure sores
2/2 (100%)
1/2 (50%)
ADLs=Activities of Daily Living; BMI=Body Mass Index; BUN=Blood urea nitrogen; CCI=Charlson Comorbidity
Index; CPS=Cognitive Performance Scale; GIC=Geriatrics Index of Comorbidity; MMSE=Mini-Mental State
Examination; MNA=Mini-Nutritional Assessment; MSQ=Mental Status Questionnaire
76
Relative strength of association of domains with mortality across individual studies
in multivariable analysis
The relative strength of association of each domain with mortality is summarized
in Table 11. For each article category, domains that had ≥50% relative strength (Panel
A) and ≥75% relative strength (Panel B) were identified, as were those achieving the
same in high quality articles alone (“relative strength” defined as the number of instances
a given domain outranked any other domain divided by the total number of instances in
which the given domain was ranked against other domains, multiplied by 100%).
General nursing home residents
Among all articles regardless of quality, measures of physical function, nutrition,
and shortness of breath had ≥75% relative strength, while disease diagnosis had ≥50%
relative strength. Among high quality articles only, nutritional measures and shortness of
breath had ≥75% relative strength, while disease diagnosis had ≥50% relative strength.
Nursing home residents with advanced dementia
In both the analysis of all articles and the analysis high quality articles only,
physical function and shortness of breath had ≥ 75% relative strength, and no other
domain had even as much as ≥50% relative strength.
Hospitalized patients: in-hospital mortality
Among all articles regardless of quality, nutrition and pressure sores were the only
domains to have ≥75% relative strength, whereas disease diagnosis had ≥50% relative
strength. Among high quality articles only, pressure sores solely had ≥75% relative
strength, while measures of disease diagnosis had ≥50% relative strength.
77
Hospitalized patients: mortality up to one year
Disease diagnosis, physical function, and nutrition all had ≥75% relative strength
among all articles regardless of quality. Among high quality articles only, physical
function and nutrition had ≥75% relative strength.
78
Table 11. Domains most strongly associated with short-term mortality in
multivariable analysis as assessed by calculations of relative strength
Panel A
cognitive
function*
disease
diagnosis
physical
function
laboratory
values
pressure
sores
nutrition
shortness
of breath
nursing
home
nursing
home:
with
advanced
dementia
hospital:
inpatient
mortality
†
†
hospital:
mortality
up to one
year
†
*All shaded boxes represent ≥50% relative strength (i.e., domains that outranked the other domains in ≥50% of instances that they
were compared in multivariable analysis); darker shadings represent that this value was maintained in a separate analysis involving
high quality articles only.
†These domains were less commonly examined in articles containing multivariable analysis, so their respective measures of relative
strength may be more subject to chance (see methods).
Panel B
cognitive
function‡
disease
diagnosis
physical
function
laboratory
values
pressure
sores
nutrition
shortness
of breath
nursing
home
nursing
home:
with
advanced
dementia
hospital:
inpatient
mortality
hospital:
mortality
up to one
year
†
†
§
†
‡All shaded boxes represent ≥75% relative strength in head-to-head comparisons with other domains; darker shadings represent that
this value was maintained in a separate analysis involving high quality articles only.
†These domains were less commonly examined in articles containing multivariable analysis, so their respective measures of relative
strength may be more subject to chance (see methods).
§Nutrition had ≥75% relative strength for high quality articles only, but not for all articles regardless of quality.
79
DISCUSSION
This systematic review of the literature identified numerous health-related
characteristics of hospitalized patients and nursing home residents significantly
associated with short-term mortality. Despite the large number of individual
characteristics, we were able to group them into a smaller number of clinically
meaningful domains: cognitive function, disease diagnosis, physical function, laboratory
values, nutrition, pressure sores, and shortness of breath. When we synthesized the
results by calculating the relative strengths of domains across studies, based on
performance compared to other domains in multivariable analysis within each study, the
emerging patterns identified the domains of health that appear to be most important for
prognostication. Among general nursing home residents, measures of nutrition and
shortness of breath were the most important, while disease diagnosis and physical
function were important to a lesser degree. Among nursing home residents with
advanced dementia, physical function and shortness of breath were the most important.
In the hospitalized elderly, disease diagnosis, nutrition, and pressure sores were the most
important for in-hospital mortality; on the other hand, for mortality up to one year
following admission, physical function and nutrition were the most important domains,
while disease diagnosis had a lesser importance.
A Domain-Based Approach to Prognostication
Although the literature describes a large number of individual health-related
characteristics associated with short-term mortality in the elderly, our review identifies a
few domains of patient health that are most strongly and consistently associated with
mortality across populations of hospitalized patients and nursing home residents. One
80
possible explanation for this consistency is that characteristics within the most important
domains may be summary measures of a patient’s health that cut across individual
disease diagnoses and other domains as well. It has been suggested that a person’s
functional status may incorporate aspects of disease burden, cognitive status, and
nutritional status.111-113 Nutritional deficits have been thought to both influence and be
influenced by disease burden and functional status.114,115 Dyspnea, commonly a marker
of cardiac and respiratory disorders, may also induce a high degree of disability.116,117 In
this way, it would seem that these domains are interrelated and broadly reflect a patient’s
state of health. As risk factors for mortality, they may reflect physiologic reserve in the
face of disease burden and other deficits. In the context of medical decisions for elderly
patients approaching the end of life, our findings are in keeping with prior arguments that
an emphasis on disease to the exclusion of other aspects of patient health may result in
inaccurate assessments, overtreatments, and mistreatments.118
The domains identified in this review as most important can be easily evaluated in
any clinical setting and do not necessarily involve calculations or specific laboratory
measurements (nutritional measures include not only albumin, but also BMI, swallowing
problems, and weight loss). Although a typical geriatric assessment would provide
information about these domains,119 the substantial lack of physicians’ evaluation of
functional and nutritional status has been shown in studies involving elderly patients in
both the primary care setting120 and hospital setting.121 Our findings argue for the utility
of incorporating these aspects of assessment into the routine clinical evaluation of the
elderly patient.
81
Rather than aiming for precise prognostic estimates, physicians could use these
measurements to easily recognize patients for whom an increased risk for mortality might
warrant a re-evaluation of the approach to care. Those patients at increased risk would be
defined by the presence of impairments in one or more of the most important domains of
health, which effectively serve as markers of vulnerability. The recognition of increased
risk for a particular patient may, for example, prompt physicians to engage in advance
care planning, to discuss hospice as a potential future option in end-of-life care, to
evaluate their willingness to undergo major or minor therapies, or simply to exercise
more caution in recommending burdensome treatments or marginally beneficial health
screening. Other potential topics for discussion include aspects of life choices, such as
financial planning and housing arrangements.59 Notably, this altered approach on the part
of the physician would not necessarily require frank discussions about prognosis with
patients.
As an additional possibility, physicians might use knowledge of the most
important domains of health to frame discussions about prognosis with patients and
families. A discussion framed around domains of health may enable patients and families
to appreciate the patient’s health in a broader context rather than focusing on a specific
diagnosis with a potentially uncertain prognostic course. Changes in these domains may
also be readily apparent to patients and families, whereas disease progression may not be
apparent without laboratory testing or imaging. Such discussions might aid patients in
their transition from being seriously ill to dying.36
This proposed alternative approach to prognostication addresses concerns
physicians have expressed that they are insufficiently prepared to prognosticate or find it
82
difficult, and that patients expect too much certainty in prognostication and might judge
them adversely if the prognosis is inaccurate.56 A domain-based method of identifying
persons at increased mortality risk would be simple for physicians to use because it
avoids complex risk calculations composed of various specific measurements that
physicians are unlikely to utilize, and it applies to elderly individuals regardless of
primary disease diagnosis. Additionally, its inability to generate specific time estimates
avoids both the problem of the inaccuracy of mortality risk calculators and the potential
for patients to overestimate the accuracy of prognostic calculations. Finally, it parallels
recent research suggesting an altered approach to hospice eligibility criteria that involves
broader patient characteristics, such as decline in functional status, quality of life, and
burden of symptoms.30 In order for such an approach to be applied to something so
specific as hospice eligibility, guidelines involving broader patient characteristics would
need to be made, and significantly more research would most likely need to be performed
to build a foundation for it.
Considering the Potential Reversibility of Domain Deficits
Identifying abnormalities in the domains of health most strongly associated with
mortality introduces the inherent challenge of deciding whether or not intervention is
likely to benefit the patient, although admittedly our review does not demonstrate a
causal relationship between any domain and mortality. On the one hand, if a patient has
an accumulating burden of these factors, interventions would seem less likely to affect
the prognosis and may have considerable adverse consequences. Nonetheless, studies
have shown that some of these factors can be individually addressed with success,
suggesting that for certain persons, these factors could be reasonable targets for
83
intervention. Several randomized control trials have shown that interventions can prevent
physical functional decline in elderly persons both in the hospital and community.122-124
Pressure sores can be effectively treated, and early prevention measures are important in
reducing their occurrence.125,126 A Cochrane review indicates that protein and energy
supplementation may reduce mortality and the risk of complications.127
The challenge for future research, then, is to determine for individual patients
whether interventions might improve their prognosis. Greater insight into the interplay
among the domains associated with mortality would likely aid in this endeavor.
Mapping, which seeks to identify the various routes through which elements can precede
or contribute to other elements, is a potential way of achieving this. Another possibility
is to examine a large cohort of elderly patients and assign to each domain a percent
contribution to mortality.128 Some of the interventional studies previously mentioned that
sought to improve patient outcomes could be used to conjecture about the likelihood of
any individual intervention having a positive effect on mortality for a patient. Ultimately,
the most definitive way to identify patients who would benefit from a multi-component
intervention is to design a randomized controlled trial and differentiate elderly subpopulations according to various outcome measures.
Limitations
One limitation to our systematic review is the heterogeneity of study cohorts.
While this issue precluded a meta-analysis, we were still able to synthesize the data in a
meaningful way by defining strict inclusion criteria regarding study populations.
Specifically, we did not include studies whose exclusion criteria eliminated patients with
characteristics potentially associated with mortality. Additionally, while our newly
84
developed method of calculating relative strength of association of domains with shortterm mortality is a helpful way of synthesizing cohort study data, there is a risk of
oversimplification in categorizing different characteristics under the heading of one
domain. Although conclusions about the usefulness of specific health-related
characteristics cannot be made based on these calculations, conclusions about the relative
importance of domains are more in keeping with this limitation. Furthermore, our first
method of analysis involving the frequency with which individual factors were associated
with mortality in bivariate and multivariable analyses provides some data comparing
characteristics within domains.
Notably, a few seminal studies on mortality risk for hospitalized patients and
nursing home residents were not appropriate for the systematic review either because the
study population included non-elderly persons or because the follow-up period was
longer than our specified limit of 1 year. The SUPPORT prognostic model was
developed in a population of adults 18 years or older with severe illness,43,129 and to our
knowledge, the model was never tested in a cohort of elderly patients. While the HELP
prognostic model was developed in a cohort of persons 80 years or older, the follow-up
period was 2 years.130 The Charlson comorbidity index, while developed and validated in
non-elderly populations,49 was assessed in several cohort studies included in this
review.77,104,105,110 Finally, the MDS-CHESS scale for nursing home residents had a 3year follow-up and included non-elderly persons.131
85
Conclusion
This review identifies several domains of particular importance in their
association with short-term mortality (defined as mortality within one year or less) in the
elderly. These domains, including physical function, nutrition, and shortness of breath,
were important when compared to other domains in results of multivariable analysis
across studies. Our findings argue for the inclusion of these domains in the general
assessment of the elderly patient, despite evidence in the literature that physicians often
do not include them in evaluations. They may be especially of value in easily identifying
elderly persons whose increased risk for short-term mortality might prompt a reevaluation of the approach to care. The challenge for future research is to identify
patients for whom interventions might improve prognosis by the reversal of domain
deficits.
86
OVERALL CONCLUSION
This thesis approaches the problem of prognostication in medicine with two
empiric studies that follow a historical introduction. The introduction traces the shift that
occurred in medicine from the first medical “tradition” to the second, the resulting
transformation of the concept of prognosis, and its decline in status in the practice of
medicine. Then a prospective cohort study illustrates the substantial limitations of the
currently accepted approach to prognostication with the specific example of physicians’
decisions to discuss hospice with seriously ill elderly patients. Finally, a systematic
review identifies the domains of patient health most strongly and consistently associated
with short-term mortality in the elderly, in an attempt to suggest an approach to
prognostication that does not seek estimates of life expectancy.
As historical writings convincingly attest, prognostication has suffered from a
tremendous lack of attention and appreciation for more than a century, in the aftermath of
a scientific revolution that dramatically advanced diagnostics and treatment options but
shifted attention from diseased individuals to individual diseases. After a long period of
virtually no discussion, a movement to improve the science of prognostication through
quantification has grown, but it has offered disappointing results in terms of limited
accuracy and lack of influence on physician-patient communication. The Medicare
Hospice Benefit requirement serves to reinforce this fallacious approach to
prognostication. Because prognostication is now largely limited to quantitative estimates
of life expectancy, physicians’ inherent uncertainty about prognosis is a barrier to
communication with patients as they approach the end of life. Prognostication, long
proclaimed an essential component in the practice of medicine, now must acknowledge
87
that certainty about life expectancy is not achievable and alter its focus to assess for
broad elements of health that indicate increased mortality risk. This is particularly
important for elderly patients, many of whom have comorbid conditions and may be
significantly harmed by available treatments, thus making irrelevant the “diagnose-andtreat” approach to medicine that had originally diminished the need for prognostic skills.
The future of prognostication is exciting, but requires a major shift in both clinical
research and clinical practice. Unfortunately, the current pursuit of the “perfect”
prognostic model has great appeal and heavy momentum behind it despite the
disappointing objective evidence. Nonetheless, in the era of an aging population with
ever-increasing complexity, it is hopeful that physicians’ desire to optimally care for their
patients will eventually restore prognostication to its proper and historical place alongside
diagnosis and treatment.
88
REFERENCES
1.
Conrad L, Neve M, Nutton V, Porter R, Wear A. The Western Medical Tradition:
800 BC to AD 1800. Cambridge: Cambridge University Press; 1995.
2.
Bynum WF, Hardy A, Jacyna S, Lawrence C, Tansey EM. The Western Medical
Tradition: 1800 to 2000. Cambridge: Cambridge University Press; 2006.
3.
Robert H. A Series of Signed Articles contributed by invitation. The Lancet
1934;223:697-8.
4.
Adams F. The Genuine Works of Hippocrates. Birmingham, AL: The Classics of
Medicine Library; 1985.
5.
Cooper G. Galen, De diebus decretoriis, from Greek into Arabic. Burlington, VT:
Ashgate; 2011.
6.
O'Boyle C. Medieval Prognosis and Astrology: A Working Edition of the
Aggregationes de crisi et creticis diebus: with Introduction and English Summary.
Cambridge: Cambridge Wellcome Unit for the History of Medicine; 1991.
7.
Wyer R. Prognostication, drawen out of the bookes of Ipocras, Avicen, and other
notable auctours of physycke. London: Robert Wyer; 1550.
8.
Alpini P. The Presages of Life and Death in Diseases. In Seven Books. London;
1746.
9.
Harvey J. Praesagium medicium, or, the prognostic signs of acute diseases.
London; 1713.
10.
Le Roy C. Oservations on the prognostic in acute diseases. 1782.
11.
Duckworth D. The Sequels of Disease. London: Longmans, Green, and Co.;
1896.
89
12.
Christakis NA. The ellipsis of prognosis in modern medical thought. Soc Sci Med
1997;44:301-15.
13.
Osler W. The Evolution of Modern Medicine. Birmingham, AL: The Classics of
Medicine Library; 1982.
14.
Harvey A., ed. Osler's Textbook Revisited. New York City: Meredith Publishing
Company; 1967.
15.
Prognosis. Journal of the American Medical Association 1901;XXXVII:915.
16.
Dock G. Prognosis; its theory and practice. Oration of medicine at the fifty-fifth
annual session of the American Medical Association at Atlantic City, June 7 to 10, 1904.
Journal of the American Medical Association 1904;XLII:1540-5.
17.
White PD. Principles and practice of prognosis, with particular reference to heart
disease. J Am Med Assoc 1953;153:75-9.
18.
Schnur S. Mortality rates in acute myocardial infarction. II. A proposed method
for measuring quantitatively severity of illness on admission to the hospital. Ann Intern
Med 1953;39:1018-25.
19.
Bleich HL. Prognosis by calculation. N Engl J Med 1971;285:1533-4.
20.
A controlled trial to improve care for seriously ill hospitalized patients. The study
to understand prognoses and preferences for outcomes and risks of treatments
(SUPPORT). The SUPPORT Principal Investigators. Jama 1995;274:1591-8.
21.
Lynn J, Teno JM, Harrell FE, Jr. Accurate prognostications of death.
Opportunities and challenges for clinicians. West J Med 1995;163:250-7.
90
22.
Christakis NA, Iwashyna TJ. Attitude and self-reported practice regarding
prognostication in a national sample of internists. Archives of Internal Medicine
1998;158:2389-95.
23.
Lynn J, Harrell F, Jr., Cohn F, Wagner D, Connors AF, Jr. Prognoses of seriously
ill hospitalized patients on the days before death: implications for patient care and public
policy. New Horiz 1997;5:56-61.
24.
Redelmeier DA, Lustig AJ. Prognostic indices in clinical practice. JAMA
2001;285:3024-5.
25.
Siontis GC, Tzoulaki I, Ioannidis JP. Predicting Death: An Empirical Evaluation
of Predictive Tools for Mortality. Arch Intern Med 2011.
26.
Minne L, Ludikhuize J, de Rooij SE, Abu-Hanna A. Characterizing predictive
models of mortality for older adults and their validation for use in clinical practice.
Journal of the American Geriatrics Society 2011;59:1110-5.
27.
Yourman LC, Lee SJ, Schonberg MA, Widera EW, Smith AK. Prognostic indices
for older adults: a systematic review. JAMA 2012;307:182-92.
28.
Gill TM. The central role of prognosis in clinical decision making. JAMA
2012;307:199-200.
29.
Harrold J, Lynn J. A Good Dying: Shaping Health Care for the Last Months of
Life. Binghamton, NY: Haworth Press; 1998.
30.
Casarett DJ, Fishman JM, Lu HL, et al. The terrible choice: re-evaluating hospice
eligibility criteria for cancer. J Clin Oncol 2009;27:953-9.
31.
Casarett DJ, Quill TE. "I'm not ready for hospice": strategies for timely and
effective hospice discussions. Annals of Internal Medicine 2007;146:443-9.
91
32.
Friedman BT, Harwood MK, Shields M. Barriers and enablers to hospice
referrals: an expert overview. Journal of palliative medicine 2002;5:73-84.
33.
Teno JM, Clarridge BR, Casey V, et al. Family perspectives on end-of-life care at
the last place of care. JAMA : the journal of the American Medical Association
2004;291:88-93.
34.
Hanson LC, Danis M, Garrett J. What is wrong with end-of-life care? Opinions of
bereaved family members. J Am Geriatr Soc 1997;45:1339-44.
35.
Cherlin E, Fried T, Prigerson HG, Schulman-Green D, Johnson-Hurzeler R,
Bradley EH. Communication between physicians and family caregivers about care at the
end of life: when do discussions occur and what is said? Journal of palliative medicine
2005;8:1176-85.
36.
Finucane TE. How gravely ill becomes dying: a key to end-of-life care. Jama
1999;282:1670-2.
37.
Bradley EH, Fried TR, Kasl SV, Cicchetti DV, Johnson-Hurzeler R, Horwitz SM.
Referral of terminally ill patients for hospice: frequency and correlates. Journal of
palliative care 2000;16:20-6.
38.
Lamont EB, Christakis NA. Physician factors in the timing of cancer patient
referral to hospice palliative care. Cancer 2002;94:2733-7.
39.
Richards J, Takeuchi LR. Factors that influence physicians' recommendation of
hospice care: An exploratory study. Journal of hospital marketing & public relations
2006;17:3-25.
92
40.
Thomas JM, O'Leary JR, Fried TR. Understanding their options: determinants of
hospice discussion for older persons with advanced illness. J Gen Intern Med
2009;24:923-8.
41.
Fried TR, Bradley EH, Towle VR, Allore H. Understanding the treatment
preferences of seriously ill patients. N Engl J Med 2002;346:1061-6.
42.
Medical guidelines for determining prognosis in selected non-cancer diseases.
The National Hospice Organization. Hosp J 1996;11:47-63.
43.
Knaus WA, Harrell FE, Jr., Lynn J, et al. The SUPPORT prognostic model.
Objective estimates of survival for seriously ill hospitalized adults. Study to understand
prognoses and preferences for outcomes and risks of treatments. Ann Intern Med
1995;122:191-203.
44.
Murphy DJ, Knaus WA, Lynn J. Study population in SUPPORT: patients (as
defined by disease categories and mortality projections), surrogates, and physicians. J
Clin Epidemiol 1990;43 Suppl:11S-28S.
45.
Lawton MP, Brody EM. Assessment of older people: self-maintaining and
instrumental activities of daily living. Gerontologist 1969;9:179-86.
46.
Inouye SK, Peduzzi PN, Robison JT, Hughes JS, Horwitz RI, Concato J.
Importance of functional measures in predicting mortality among older hospitalized
patients. Jama 1998;279:1187-93.
47.
Royall DR, Mahurin RK, Gray KF. Bedside assessment of executive cognitive
impairment: the executive interview. J Am Geriatr Soc 1992;40:1221-6.
48.
Pfeiffer E. A short portable mental status questionnaire for the assessment of
organic brain deficit in elderly patients. J Am Geriatr Soc 1975;23:433-41.
93
49.
Charlson ME, Pompei P, Ales KL, MacKenzie CR. A new method of classifying
prognostic comorbidity in longitudinal studies: development and validation. J Chronic
Dis 1987;40:373-83.
50.
Katz S, Ford AB, Moskowitz RW, Jackson BA, Jaffe MW. Studies of Illness in
the Aged. the Index of Adl: a Standardized Measure of Biological and Psychosocial
Function. Jama 1963;185:914-9.
51.
Walke LM, Gallo WT, Tinetti ME, Fried TR. The burden of symptoms among
community-dwelling older persons with advanced chronic disease. Arch Intern Med
2004;164:2321-4.
52.
Pearlin LI, Lieberman MA, Menaghan EG, Mullan JT. The stress process. J
Health Soc Behav 1981;22:337-56.
53.
Byock IR, Merriman MP. Measuring quality of life for patients with terminal
illness: the Missoula-VITAS quality of life index. Palliat Med 1998;12:231-44.
54.
Weggel JM. Barriers to the physician decision to offer hospice as an option for
terminal care. WMJ : official publication of the State Medical Society of Wisconsin
1999;98:49-53.
55.
Fox E, Landrum-McNiff K, Zhong Z, Dawson NV, Wu AW, Lynn J. Evaluation
of prognostic criteria for determining hospice eligibility in patients with advanced lung,
heart, or liver disease. SUPPORT Investigators. Study to Understand Prognoses and
Preferences for Outcomes and Risks of Treatments. JAMA 1999;282:1638-45.
56.
Christakis NA, Iwashyna TJ. Attitude and self-reported practice regarding
prognostication in a national sample of internists. Arch Intern Med 1998;158:2389-95.
94
57.
Daugherty CK, Hlubocky FJ. What are terminally ill cancer patients told about
their expected deaths? A study of cancer physicians' self-reports of prognosis disclosure.
J Clin Oncol 2008;26:5988-93.
58.
Quill TE. Perspectives on care at the close of life. Initiating end-of-life
discussions with seriously ill patients: addressing the "elephant in the room". Jama
2000;284:2502-7.
59.
Smith AK, Williams BA, Lo B. Discussing overall prognosis with the very
elderly. N Engl J Med 2011;365:2149-51.
60.
Campbell SE, Seymour DG, Primrose WR. A systematic literature review of
factors affecting outcome in older medical patients admitted to hospital. Age Ageing
2004;33:110-5.
61.
Hayden JA, Cote P, Bombardier C. Evaluation of the quality of prognosis studies
in systematic reviews. Ann Intern Med 2006;144:427-37.
62.
Rothwell PM. External validity of randomised controlled trials: "to whom do the
results of this trial apply?". Lancet 2005;365:82-93.
63.
Barca ML, Engedal K, Laks J, et al. A 12 months follow-up study of depression
among nursing-home patients in Norway. J Affect Disord 2010;120:141-8.
64.
Flacker JM, Kiely DK. A practical approach to identifying mortality-related
factors in established long-term care residents. J Am Geriatr Soc 1998;46:1012-5.
65.
Flacker JM, Kiely DK, Flacker JM, Kiely DK. Mortality-related factors and 1-
year survival in nursing home residents. J Am Geriatr Soc 2003;51:213-21.
95
66.
Grabowski DC, Campbell CM, Ellis JE, Grabowski DC, Campbell CM, Ellis JE.
Obesity and mortality in elderly nursing home residents. J Gerontol A Biol Sci Med Sci
2005;60:1184-9.
67.
Kiely DK, Flacker JM. Resident characteristics associated with mortality in long-
term care nursing homes: is there a gender difference? J Am Med Dir Assoc 2000;1:8-13.
68.
Kiely DK, Flacker JM. Common and gender specific factors associated with one-
year mortality in nursing home residents. J Am Med Dir Assoc 2002;3:302-9.
69.
Mooradian AD, Reed RL, Osterweil D, Scuderi P. Detectable serum levels of
tumor necrosis factor alpha may predict early mortality in elderly institutionalized
patients. J Am Geriatr Soc 1991;39:891-4.
70.
Perls TT, Morris JN, Ooi WL, Lipsitz LA. The relationship between age, gender
and cognitive performance in the very old: the effect of selective survival. J Am Geriatr
Soc 1993;41:1193-201.
71.
Tsai AC, Ku PY, Tsai AC, Ku P-Y. Population-specific Mini Nutritional
Assessment effectively predicts the nutritional state and follow-up mortality of
institutionalized elderly Taiwanese regardless of cognitive status. Br J Nutr
2008;100:152-8.
72.
Mitchell SL, Kiely DK, Hamel MB, et al. Estimating prognosis for nursing home
residents with advanced dementia. Jama 2004;291:2734-40.
73.
Mitchell SL, Miller SC, Teno JM, Davis RB, Shaffer ML. The advanced dementia
prognostic tool: a risk score to estimate survival in nursing home residents with advanced
dementia. J Pain Symptom Manage 2010;40:639-51.
96
74.
Alarcon T, Barcena A, Gonzalez-Montalvo JI, Penalosa C, Salgado A. Factors
predictive of outcome on admission to an acute geriatric ward. Age & Ageing
1999;28:429-32.
75.
Boyd CM, Landefeld CS, Counsell SR, et al. Recovery of activities of daily living
in older adults after hospitalization for acute medical illness. J Am Geriatr Soc
2008;56:2171-9.
76.
Buurman BM, van Munster BC, Korevaar JC, et al. Prognostication in acutely
admitted older patients by nurses and physicians. Journal of General Internal Medicine
2008;23:1883-9.
77.
Covinsky KE, Justice AC, Rosenthal GE, Palmer RM, Landefeld CS. Measuring
prognosis and case mix in hospitalized elders. The importance of functional status.
Journal of General Internal Medicine 1997;12:203-8.
78.
Desai MM, Bogardus ST, Jr., Williams CS, et al. Development and validation of a
risk-adjustment index for older patients: the high-risk diagnoses for the elderly scale. J
Am Geriatr Soc 2002;50:474-81.
79.
Drame M, Jovenin N, Novella JL, et al. Predicting early mortality among elderly
patients hospitalised in medical wards via emergency department: the SAFES cohort
study. J Nutr Health Aging 2008;12:599-604.
80.
Eeles EM, Hubbard RE, White SV, O'Mahony MS, Savva GM, Bayer AJ.
Hospital use, institutionalisation and mortality associated with delirium. Age Ageing
2010;39:470-5.
81.
Flodin L, Svensson S, Cederholm T. Body mass index as a predictor of 1 year
mortality in geriatric patients. Clin Nutr 2000;19:121-5.
97
82.
Gonzalez M, Martinez G, Calderon J, et al. Impact of delirium on short-term
mortality in elderly inpatients: a prospective cohort study. Psychosomatics 2009;50:2348.
83.
Inouye SK, Bogardus ST, Jr., Vitagliano G, et al. Burden of illness score for
elderly persons: risk adjustment incorporating the cumulative impact of diseases,
physiologic abnormalities, and functional impairments.[Erratum appears in Med Care.
2003 Mar;41(3):446]. Med Care 2003;41:70-83.
84.
Laurila JV, Pitkala KH, Strandberg TE, et al. Impact of different diagnostic
criteria on prognosis of delirium: a prospective study. Dement Geriatr Cogn Disord
2004;18:240-4.
85.
Narain P, Rubenstein LZ, Wieland GD, et al. Predictors of immediate and 6-
month outcomes in hospitalized elderly patients. The importance of functional status. J
Am Geriatr Soc 1988;36:775-83.
86.
Persson MD, Brismar KE, Katzarski KS, et al. Nutritional status using mini
nutritional assessment and subjective global assessment predict mortality in geriatric
patients. J Am Geriatr Soc 2002;50:1996-2002.
87.
Pilotto A, Ferrucci L, Franceschi M, et al. Development and validation of a
multidimensional prognostic index for one-year mortality from comprehensive geriatric
assessment in hospitalized older patients. Rejuvenation Res 2008;11:151-61.
88.
Pitkala KH, Laurila JV, Strandberg TE, et al. Prognostic significance of delirium
in frail older people. Dement Geriatr Cogn Disord 2005;19:158-63.
98
89.
van Doorn C, Bogardus ST, Williams CS, Concato J, Towle VR, Inouye SK. Risk
adjustment for older hospitalized persons: a comparison of two methods of data
collection for the Charlson index. J Clin Epidemiol 2001;54:694-701.
90.
Walter LC, Brand RJ, Counsell SR, et al. Development and validation of a
prognostic index for 1-year mortality in older adults after hospitalization. Jama
2001;285:2987-94.
91.
Abizanda P, Navarro JL, Romero L, Leon M, Sanchez-Jurado PM, Dominguez L.
Upper extremity function, an independent predictor of adverse events in hospitalized
elderly. Gerontology 2007;53:267-73.
92.
Agarwal N, Acevedo F, Leighton LS, Cayten CG, Pitchumoni CS. Predictive
ability of various nutritional variables for mortality in elderly people. Am J Clin Nutr
1988;48:1173-8.
93.
Bienia R, Ratcliff S, Barbour GL, Kummer M. Malnutrition in the hospitalized
geriatric patient. J Am Geriatr Soc 1982;30:433-6.
94.
Gazzotti C, Albert A, Pepinster A, Petermans J. Clinical usefulness of the mini
nutritional assessment (MNA) scale in geriatric medicine. J Nutr Health Aging
2000;4:176-81.
95.
Incalzi RA, Gemma A, Capparella O, et al. Predicting mortality and length of stay
of geriatric patients in an acute care general hospital. J Gerontol 1992;47:M35-9.
96.
Antonelli Incalzi R, Landi F, Cipriani L, et al. Nutritional assessment: a primary
component of multidimensional geriatric assessment in the acute care setting. J Am
Geriatr Soc 1996;44:166-74.
99
97.
Incalzi RA, Capparella O, Gemma A, et al. The interaction between age and
comorbidity contributes to predicting the mortality of geriatric patients in the acute-care
hospital. J Intern Med 1997;242:291-8.
98.
Iwata M, Kuzuya M, Kitagawa Y, et al. Prognostic value of serum albumin
combined with serum C-reactive protein levels in older hospitalized patients: continuing
importance of serum albumin. Aging Clin Exp Res 2006;18:307-11.
99.
Jonsson PV, Noro A, Finne-Soveri H, et al. Admission profile is predictive of
outcome in acute hospital care. Aging Clin Exp Res 2008;20:533-9.
100.
Marengoni A, Cossi S, De Martinis M, et al. Adverse outcomes in older
hospitalized patients: the role of multidimensional geriatric assessment. Aging Clin Exp
Res 2003;15:32-7.
101.
Marengoni A, Petroboni B, Casella S, et al. Total lymphocyte count and in-
hospital mortality in older persons with multimorbidity. Aging Clin Exp Res
2008;20:290-6.
102.
O'Keeffe S, Lavan J. The prognostic significance of delirium in older hospital
patients. J Am Geriatr Soc 1997;45:174-8.
103.
Pompei P, Foreman M, Rudberg MA, Inouye SK, Braund V, Cassel CK. Delirium
in hospitalized older persons: outcomes and predictors. J Am Geriatr Soc 1994;42:80915.
104.
Ponzetto M, Maero B, Maina P, et al. Risk factors for early and late mortality in
hospitalized older patients: the continuing importance of functional status. J Gerontol A
Biol Sci Med Sci 2003;58:1049-54.
100
105.
Sampson EL, Blanchard MR, Jones L, et al. Dementia in the acute hospital:
prospective cohort study of prevalence and mortality. Br J Psychiatry 2009;195:61-6.
106.
Sonnenblick M, Raveh D, Gratch L, Yinnon A. Clinical and demographic
characteristics of elderly patients hospitalised in an internal medicine department in
Israel. Int J Clin Pract 2007;61:247-54.
107.
Stratton RJ, King CL, Stroud MA, et al. 'Malnutrition Universal Screening Tool'
predicts mortality and length of hospital stay in acutely ill elderly. Br J Nutr
2006;95:325-30.
108.
Terzian C, Frye EB, Piotrowski ZH. Admission hyponatremia in the elderly:
factors influencing prognosis. Journal of General Internal Medicine 1994;9:89-91.
109.
Zafrir B, Laor A, Bitterman H, Zafrir B, Laor A, Bitterman H. Nonagenarians in
internal medicine: characteristics, outcomes and predictors for in-hospital and postdischarge mortality. Israel Medical Association Journal: Imaj;12:10-5.
110.
Zekry D, Loures Valle BH, Lardi C, et al. Geriatrics index of comorbidity was the
most accurate predictor of death in geriatric hospital among six comorbidity scores. J
Clin Epidemiol 2010;63:1036-44.
111.
Studenski S, Perera S, Wallace D, et al. Physical performance measures in the
clinical setting. J Am Geriatr Soc 2003;51:314-22.
112.
Applegate WB, Blass JP, Williams TF. Instruments for the functional assessment
of older patients. N Engl J Med 1990;322:1207-14.
113.
Guralnik JM, Branch LG, Cummings SR, Curb JD. Physical performance
measures in aging research. J Gerontol 1989;44:M141-6.
101
114.
Chen CC, Schilling LS, Lyder CH. A concept analysis of malnutrition in the
elderly. J Adv Nurs 2001;36:131-42.
115.
de Groot CP, van Staveren WA. Nutritional concerns, health and survival in old
age. Biogerontology 2010;11:597-602.
116.
Tessier JF, Nejjari C, Letenneur L, et al. Dyspnea and 8-year mortality among
elderly men and women: the PAQUID cohort study. Eur J Epidemiol 2001;17:223-9.
117.
Huijnen B, van der Horst F, van Amelsvoort L, et al. Dyspnea in elderly family
practice patients. Occurrence, severity, quality of life and mortality over an 8-year period.
Fam Pract 2006;23:34-9.
118.
Tinetti ME, Fried T. The end of the disease era. Am J Med 2004;116:179-85.
119.
Geriatric assessment methods for clinical decisionmaking. Natl Inst Health
Consens Dev Conf Consens Statement 1987;6:1-8.
120.
Wenger NS, Solomon DH, Roth CP, et al. The quality of medical care provided to
vulnerable community-dwelling older patients. Ann Intern Med 2003;139:740-7.
121.
Arora VM, Johnson M, Olson J, et al. Using assessing care of vulnerable elders
quality indicators to measure quality of hospital care for vulnerable elders. J Am Geriatr
Soc 2007;55:1705-11.
122.
Gill TM, Baker DI, Gottschalk M, Peduzzi PN, Allore H, Byers A. A program to
prevent functional decline in physically frail, elderly persons who live at home. N Engl J
Med 2002;347:1068-74.
123.
Counsell SR, Holder CM, Liebenauer LL, et al. Effects of a multicomponent
intervention on functional outcomes and process of care in hospitalized older patients: a
102
randomized controlled trial of Acute Care for Elders (ACE) in a community hospital. J
Am Geriatr Soc 2000;48:1572-81.
124.
Landefeld CS, Palmer RM, Kresevic DM, Fortinsky RH, Kowal J. A randomized
trial of care in a hospital medical unit especially designed to improve the functional
outcomes of acutely ill older patients. N Engl J Med 1995;332:1338-44.
125.
Brandeis GH, Morris JN, Nash DJ, Lipsitz LA. The epidemiology and natural
history of pressure ulcers in elderly nursing home residents. Jama 1990;264:2905-9.
126.
Reddy M, Gill SS, Rochon PA. Preventing pressure ulcers: a systematic review.
Jama 2006;296:974-84.
127.
Milne AC, Potter J, Vivanti A, Avenell A. Protein and energy supplementation in
elderly people at risk from malnutrition. Cochrane Database Syst Rev 2009:CD003288.
128.
Tinetti ME, McAvay G, Chang SS, et al. Effect of Chronic Disease–Related
Symptoms and Impairments on Universal Health Outcomes in Older Adults. Journal of
the American Geriatrics Society 2011;59:1618-27.
129.
Hamel MB, Davis RB, Teno JM, et al. Older age, aggressiveness of care, and
survival for seriously ill, hospitalized adults. SUPPORT Investigators. Study to
Understand Prognoses and Preferences for Outcomes and Risks of Treatments. Ann
Intern Med 1999;131:721-8.
130.
Teno JM, Harrell FE, Jr., Knaus W, et al. Prediction of survival for older
hospitalized patients: the HELP survival model. Hospitalized Elderly Longitudinal
Project. J Am Geriatr Soc 2000;48:S16-24.
131.
Hirdes JP, Frijters DH, Teare GF. The MDS-CHESS scale: a new measure to
predict mortality in institutionalized older people. J Am Geriatr Soc 2003;51:96-100.